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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 85 by nigel,
Sat Feb 24 21:41:13 2007 UTC
+revision 905 by zherczeg,
Mon Jan 23 19:26:03 2012 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2005 University of Cambridge
+            Copyright (c) 1997-2012 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 42 
 POSSIBILITY OF SUCH DAMAGE.
  supporting internal functions that are not used by other modules. */
+ #ifdef HAVE_CONFIG_H
+ #include "config.h"
+ #endif
+ #define NLBLOCK cd             /* Block containing newline information */
+ #define PSSTART start_pattern  /* Field containing processed string start */
+ #define PSEND   end_pattern    /* Field containing processed string end */
  #include "pcre_internal.h"
- /* When DEBUG is defined, we need the pcre_printint() function, which is also
+ /* When PCRE_DEBUG is defined, we need the pcre(16)_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
+ COMPILE_PCREx macro will already be appropriately set. */
- #ifdef DEBUG
+ #ifdef PCRE_DEBUG
- #include "pcre_printint.src"
+ /* 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))
+ /* Maximum length value to check against when making sure that the integer that
+ holds the compiled pattern length does not overflow. We make it a bit less than
+ INT_MAX to allow for adding in group terminating bytes, so that we don't have
+ to check them every time. */
+ #define OFLOW_MAX (INT_MAX - 20)
  /*************************************************
  *      Code parameters and static tables         *
  *************************************************/
- /* Maximum number of items on the nested bracket stacks at compile time. This
+ /* This value specifies the size of stack workspace that is used during the
- applies to the nesting of all kinds of parentheses. It does not limit
+ first pre-compile phase that determines how much memory is required. The regex
- un-nested, non-capturing parentheses. This number can be made bigger if
+ is partly compiled into this space, but the compiled parts are discarded as
- necessary - it is used to dimension one int and one unsigned char vector at
+ soon as they can be, so that hopefully there will never be an overrun. The code
- compile time. */
+ does, however, check for an overrun. The largest amount I've seen used is 218,
+ so this number is very generous.
+ The same workspace is used during the second, actual compile phase for
+ remembering forward references to groups so that they can be filled in at the
+ end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
+ is 4 there is plenty of room 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)
+ /* 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 BRASTACK_SIZE 200
+ #define REQ_CASELESS   0x10000000l      /* Indicates caselessness */
+ #define REQ_VARY       0x20000000l      /* Reqchar followed non-literal item */
+ /* 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. */
- #if !EBCDIC   /* This is the "normal" table for ASCII systems */
+ #ifndef EBCDIC
+ /* This is the "normal" table for ASCII systems or for EBCDIC systems running
+ in UTF-8 mode. */
  static const short int escapes[] = {
-,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */
+,                       0,
-,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */
+,                       0,
-    '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */
+,                       0,
-,      0,      0,      0,      0,      0,      0,      0,   /* H - O */
+,                       0,
- -ESC_P, -ESC_Q,      0, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */
+,                       0,
- -ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */
+      CHAR_COLON,              CHAR_SEMICOLON,
-    '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */
+      CHAR_LESS_THAN_SIGN,     CHAR_EQUALS_SIGN,
-,      0,      0,      0,      0,      0,  ESC_n,      0,   /* h - o */
+      CHAR_GREATER_THAN_SIGN,  CHAR_QUESTION_MARK,
- -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0,      0, -ESC_w,   /* p - w */
+      CHAR_COMMERCIAL_AT,      -ESC_A,
-,      0, -ESC_z                                            /* x - z */
+      -ESC_B,                  -ESC_C,
+      -ESC_D,                  -ESC_E,
+,                       -ESC_G,
+      -ESC_H,                  0,
+,                       -ESC_K,
+,                       0,
+      -ESC_N,                  0,
+      -ESC_P,                  -ESC_Q,
+      -ESC_R,                  -ESC_S,
+,                       0,
+      -ESC_V,                  -ESC_W,
+      -ESC_X,                  0,
+      -ESC_Z,                  CHAR_LEFT_SQUARE_BRACKET,
+      CHAR_BACKSLASH,          CHAR_RIGHT_SQUARE_BRACKET,
+      CHAR_CIRCUMFLEX_ACCENT,  CHAR_UNDERSCORE,
+      CHAR_GRAVE_ACCENT,       7,
+      -ESC_b,                  0,
+      -ESC_d,                  ESC_e,
+      ESC_f,                   0,
+      -ESC_h,                  0,
+,                       -ESC_k,
+,                       0,
+      ESC_n,                   0,
+      -ESC_p,                  0,
+      ESC_r,                   -ESC_s,
+      ESC_tee,                 0,
+      -ESC_v,                  -ESC_w,
+,                       0,
+      -ESC_z
  };
- #else         /* This is the "abnormal" table for EBCDIC systems */
+ #else
+ /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
  static const short int escapes[] = {
  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
-Line 96 
 static const short int escapes[] = {
+Line 181 
 static const short int escapes[] = {
  /*  70 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',
  /*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,
- /*  88 */     0,     0,      0,     '{',      0,     0,      0,      0,
+ /*  88 */-ESC_h,     0,      0,     '{',      0,     0,      0,      0,
- /*  90 */     0,     0,      0,     'l',      0, ESC_n,      0, -ESC_p,
+ /*  90 */     0,     0, -ESC_k,     'l',      0, ESC_n,      0, -ESC_p,
  /*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,
- /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,     0, -ESC_w,      0,
+ /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,-ESC_v, -ESC_w,      0,
  /*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,
  /*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
- /*  C8 */     0,     0,      0,       0,      0,     0,      0,      0,
+ /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
- /*  D0 */   '}',     0,      0,       0,      0,     0,      0, -ESC_P,
+ /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
- /*  D8 */-ESC_Q,     0,      0,       0,      0,     0,      0,      0,
+ /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
- /*  E0 */  '\\',     0, -ESC_S,       0,      0,     0, -ESC_W, -ESC_X,
+ /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
  /*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,
  /*  F8 */     0,     0,      0,       0,      0,     0,      0,      0
-Line 115 
 static const short int escapes[] = {
+Line 200 
 static const short int escapes[] = {
  #endif
- /* Tables of names of POSIX character classes and their lengths. The list is
+ /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
- terminated by a zero length entry. The first three must be alpha, upper, lower,
+ searched linearly. Put all the names into a single string, in order to reduce
- as this is assumed for handling case independence. */
+ the number of relocations when a shared library is dynamically linked. The
+ string is built from string macros so that it works in UTF-8 mode on EBCDIC
- static const char *const posix_names[] = {
+ platforms. */
-   "alpha", "lower", "upper",
-   "alnum", "ascii", "blank", "cntrl", "digit", "graph",
+ typedef struct verbitem {
-   "print", "punct", "space", "word",  "xdigit" };
+   int   len;                 /* Length of verb name */
+   int   op;                  /* Op when no arg, or -1 if arg mandatory */
+   int   op_arg;              /* Op when arg present, or -1 if not allowed */
+ } verbitem;
+ static const char verbnames[] =
+   "\0"                       /* Empty name is a shorthand for MARK */
+   STRING_MARK0
+   STRING_ACCEPT0
+   STRING_COMMIT0
+   STRING_F0
+   STRING_FAIL0
+   STRING_PRUNE0
+   STRING_SKIP0
+   STRING_THEN;
+ static const verbitem verbs[] = {
+   { 0, -1,        OP_MARK },
+   { 4, -1,        OP_MARK },
+   { 6, OP_ACCEPT, -1 },
+   { 6, OP_COMMIT, -1 },
+   { 1, OP_FAIL,   -1 },
+   { 4, OP_FAIL,   -1 },
+   { 5, OP_PRUNE,  OP_PRUNE_ARG },
+   { 4, OP_SKIP,   OP_SKIP_ARG  },
+   { 4, OP_THEN,   OP_THEN_ARG  }
+ };
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
+ /* Tables of names of POSIX character classes and their lengths. The names are
+ now all in a single string, to reduce the number of relocations when a shared
+ library is dynamically loaded. The list of lengths is terminated by a zero
+ length entry. The first three must be alpha, lower, upper, as this is assumed
+ for handling case independence. */
+ static const char posix_names[] =
+   STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
+   STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
+   STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
+   STRING_word0  STRING_xdigit;
- static const uschar posix_name_lengths[] = {
+ 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; up to three may be combined
+ /* Table of class bit maps for each POSIX class. Each class is formed from a
- to form the class. The table for [:blank:] is dynamically modified to remove
+ base map, with an optional addition or removal of another map. Then, for some
- the vertical space characters. */
+ classes, there is some additional tweaking: for [:blank:] the vertical space
+ characters are removed, and for [:alpha:] and [:alnum:] the underscore
+ character is removed. The triples in the table consist of the base map offset,
+ second map offset or -1 if no second map, and a non-negative value for map
+ addition or a negative value for map subtraction (if there are two maps). The
+ absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
+ remove vertical space characters, 2 => remove underscore. */
  static const int posix_class_maps[] = {
-   cbit_lower, cbit_upper, -1,             /* alpha */
+   cbit_word,  cbit_digit, -2,             /* alpha */
-   cbit_lower, -1,         -1,             /* lower */
+   cbit_lower, -1,          0,             /* lower */
-   cbit_upper, -1,         -1,             /* upper */
+   cbit_upper, -1,          0,             /* upper */
-   cbit_digit, cbit_lower, cbit_upper,     /* alnum */
+   cbit_word,  -1,          2,             /* alnum - word without underscore */
-   cbit_print, cbit_cntrl, -1,             /* ascii */
+   cbit_print, cbit_cntrl,  0,             /* ascii */
-   cbit_space, -1,         -1,             /* blank - a GNU extension */
+   cbit_space, -1,          1,             /* blank - a GNU extension */
-   cbit_cntrl, -1,         -1,             /* cntrl */
+   cbit_cntrl, -1,          0,             /* cntrl */
-   cbit_digit, -1,         -1,             /* digit */
+   cbit_digit, -1,          0,             /* digit */
-   cbit_graph, -1,         -1,             /* graph */
+   cbit_graph, -1,          0,             /* graph */
-   cbit_print, -1,         -1,             /* print */
+   cbit_print, -1,          0,             /* print */
-   cbit_punct, -1,         -1,             /* punct */
+   cbit_punct, -1,          0,             /* punct */
-   cbit_space, -1,         -1,             /* space */
+   cbit_space, -1,          0,             /* space */
-   cbit_word,  -1,         -1,             /* word - a Perl extension */
+   cbit_word,  -1,          0,             /* word - a Perl extension */
-   cbit_xdigit,-1,         -1              /* xdigit */
+   cbit_xdigit,-1,          0              /* xdigit */
  };
+ /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
+ substitutes must be in the order of the names, defined above, and there are
+ both positive and negative cases. NULL means no substitute. */
- /* The texts of compile-time error messages. These are "char *" because they
+ #ifdef SUPPORT_UCP
- are passed to the outside world. */
+ static const 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
- static const char *error_texts[] = {
+ #define STRING(a)  # a
-   "no error",
+ #define XSTRING(s) STRING(s)
-   "\\ at end of pattern",
-   "\\c at end of pattern",
+ /* The texts of compile-time error messages. These are "char *" because they
-   "unrecognized character follows \\",
+ are passed to the outside world. Do not ever re-use any error number, because
-   "numbers out of order in {} quantifier",
+ they are documented. Always add a new error instead. Messages marked DEAD below
+ are no longer used. This used to be a table of strings, but in order to reduce
+ the number of relocations needed when a shared library is loaded dynamically,
+ it is now one long string. We cannot use a table of offsets, because the
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
+ simply count through to the one we want - this isn't a performance issue
+ because these strings are used only when there is a compilation error.
+ Each substring ends with \0 to insert a null character. This includes the final
+ substring, so that the whole string ends with \0\0, which can be detected when
+ counting through. */
+ static const char error_texts[] =
+   "no error\0"
+   "\\ at end of pattern\0"
+   "\\c at end of pattern\0"
+   "unrecognized character follows \\\0"
+   "numbers out of order in {} quantifier\0"
    /* 5 */
-   "number too big in {} quantifier",
+   "number too big in {} quantifier\0"
-   "missing terminating ] for character class",
+   "missing terminating ] for character class\0"
-   "invalid escape sequence in character class",
+   "invalid escape sequence in character class\0"
-   "range out of order in character class",
+   "range out of order in character class\0"
-   "nothing to repeat",
+   "nothing to repeat\0"
    /* 10 */
-   "operand of unlimited repeat could match the empty string",
+   "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
-   "internal error: unexpected repeat",
+   "internal error: unexpected repeat\0"
-   "unrecognized character after (?",
+   "unrecognized character after (? or (?-\0"
-   "POSIX named classes are supported only within a class",
+   "POSIX named classes are supported only within a class\0"
-   "missing )",
+   "missing )\0"
    /* 15 */
-   "reference to non-existent subpattern",
+   "reference to non-existent subpattern\0"
-   "erroffset passed as NULL",
+   "erroffset passed as NULL\0"
-   "unknown option bit(s) set",
+   "unknown option bit(s) set\0"
-   "missing ) after comment",
+   "missing ) after comment\0"
-   "parentheses nested too deeply",
+   "parentheses nested too deeply\0"  /** DEAD **/
    /* 20 */
-   "regular expression too large",
+   "regular expression is too large\0"
-   "failed to get memory",
+   "failed to get memory\0"
-   "unmatched parentheses",
+   "unmatched parentheses\0"
-   "internal error: code overflow",
+   "internal error: code overflow\0"
-   "unrecognized character after (?<",
+   "unrecognized character after (?<\0"
    /* 25 */
-   "lookbehind assertion is not fixed length",
+   "lookbehind assertion is not fixed length\0"
-   "malformed number after (?(",
+   "malformed number or name after (?(\0"
-   "conditional group contains more than two branches",
+   "conditional group contains more than two branches\0"
-   "assertion expected after (?(",
+   "assertion expected after (?(\0"
-   "(?R or (?digits must be followed by )",
+   "(?R or (?[+-]digits must be followed by )\0"
    /* 30 */
-   "unknown POSIX class name",
+   "unknown POSIX class name\0"
-   "POSIX collating elements are not supported",
+   "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support",
+   "this version of PCRE is compiled without UTF support\0"
-   "spare error",
+   "spare error\0"  /** DEAD **/
-   "character value in \\x{...} sequence is too large",
+   "character value in \\x{...} sequence is too large\0"
    /* 35 */
-   "invalid condition (?(0)",
+   "invalid condition (?(0)\0"
-   "\\C not allowed in lookbehind assertion",
+   "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
+   "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
-   "number after (?C is > 255",
+   "number after (?C is > 255\0"
-   "closing ) for (?C expected",
+   "closing ) for (?C expected\0"
    /* 40 */
-   "recursive call could loop indefinitely",
+   "recursive call could loop indefinitely\0"
-   "unrecognized character after (?P",
+   "unrecognized character after (?P\0"
-   "syntax error after (?P",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named groups have the same name",
+   "two named subpatterns have the same name\0"
-   "invalid UTF-8 string",
+   "invalid UTF-8 string\0"
    /* 45 */
-   "support for \\P, \\p, and \\X has not been compiled",
+   "support for \\P, \\p, and \\X has not been compiled\0"
-   "malformed \\P or \\p sequence",
+   "malformed \\P or \\p sequence\0"
-   "unknown property name after \\P or \\p"
+   "unknown property name after \\P or \\p\0"
- };
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
+   /* 50 */
+   "repeated subpattern is too long\0"    /** DEAD **/
+   "octal value is greater than \\377 in 8-bit non-UTF-8 mode\0"
+   "internal error: overran compiling workspace\0"
+   "internal error: previously-checked referenced subpattern not found\0"
+   "DEFINE group contains more than one branch\0"
+   /* 55 */
+   "repeating a DEFINE group is not allowed\0"  /** DEAD **/
+   "inconsistent NEWLINE options\0"
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
+   "a numbered reference must not be zero\0"
+   "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
+   /* 60 */
+   "(*VERB) not recognized\0"
+   "number is too big\0"
+   "subpattern name expected\0"
+   "digit expected after (?+\0"
+   "] is an invalid data character in JavaScript compatibility mode\0"
+   /* 65 */
+   "different names for subpatterns of the same number are not allowed\0"
+   "(*MARK) must have an argument\0"
+   "this version of PCRE is not compiled with 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"
+   ;
  /* Table to identify digits and hex digits. This is used when compiling
  patterns. Note that the tables in chartables are dependent on the locale, and
-Line 229 
 For convenience, we use the same bit def
+Line 507 
 For convenience, we use the same bit def
  Then we can use ctype_digit and ctype_xdigit in the code. */
- #if !EBCDIC    /* This is the "normal" case, for ASCII systems */
+ /* 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 265 
 static const unsigned char digitab[] =
+Line 553 
 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 279 
 static const unsigned char digitab[] =
+Line 570 
 static const unsigned char digitab[] =
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
-x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- �     */
+x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
-Line 301 
 static const unsigned char digitab[] =
+Line 592 
 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 313 
 static const unsigned char ebcdic_charta
+Line 604 
 static const unsigned char ebcdic_charta
 x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
 x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
-x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- �  */
+x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
 x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
 x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
-Line 340 
 static const unsigned char ebcdic_charta
+Line 631 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,
+   compile_regex(int, pcre_uchar **, const pcre_uchar **, int *, BOOL, BOOL, int, int,
-     int *, int *, branch_chain *, compile_data *);
+     int *, int *, branch_chain *, compile_data *, int *);
+ /*************************************************
+ *            Find an error text                  *
+ *************************************************/
+ /* The error texts are now all in one long string, to save on relocations. As
+ some of the text is of unknown length, we can't use a table of offsets.
+ Instead, just count through the strings. This is not a performance issue
+ because it happens only when there has been a compilation error.
+ Argument:   the error number
+ Returns:    pointer to the error string
+ */
+ static const char *
+ find_error_text(int n)
+ {
+ const char *s = error_texts;
+ for (; n > 0; n--)
+   {
+   while (*s++ != 0) {};
+   if (*s == 0) return "Error text not found (please report)";
+   }
+ return s;
+ }
+ /*************************************************
+ *           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
+ is_counted_repeat(const pcre_uchar *p)
+ {
+ 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 351 
 static BOOL
+Line 740 
 static BOOL
  /* This function is called when a \ has been encountered. It either returns a
  positive value for a simple escape such as \n, or a negative value which
- encodes one of the more complicated things such as \d. When UTF-8 is enabled,
+ encodes one of the more complicated things such as \d. A backreference to group
- a positive value greater than 255 may be returned. On entry, ptr is pointing at
+ n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
- the \. On exit, it is on the final character of the escape sequence.
+ UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
+ ptr is pointing at the \. On exit, it is on the final character of the escape
+ sequence.
  Arguments:
    ptrptr         points to the pattern position pointer
-Line 364 
 Arguments:
+Line 755 
 Arguments:
  Returns:         zero or positive => a data character
                   negative => a special escape sequence
-                  on error, errorptr is set
+                  on error, errorcodeptr is set
  */
  static int
- check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
+ check_escape(const pcre_uchar **ptrptr, int *errorcodeptr, int bracount,
    int options, BOOL isclass)
  {
- const uschar *ptr = *ptrptr;
+ /* PCRE_UTF16 has the same value as PCRE_UTF8. */
- int c, i;
+ BOOL utf = (options & PCRE_UTF8) != 0;
+ const pcre_uchar *ptr = *ptrptr + 1;
+ pcre_int32 c;
+ 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. */
- c = *(++ptr);
  if (c == 0) *errorcodeptr = ERR1;
- /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
+ /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
- a table. A non-zero result is something that can be returned immediately.
+ in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphameric */
+ /* 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) c = i;
- #else          /* EBCDIC coding */
- else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */
+ #else           /* EBCDIC coding */
+ /* Not alphanumeric */
+ else if (c < 'a' || (!MAX_255(c) || (ebcdic_chartab[c] & 0x0E) == 0)) {}
  else if ((i = escapes[c - 0x48]) != 0)  c = i;
  #endif
-Line 396 
 else if ((i = escapes[c - 0x48]) != 0)
+Line 794 
 else if ((i = escapes[c - 0x48]) != 0)
  else
    {
-   const uschar *oldptr;
+   const pcre_uchar *oldptr;
+   BOOL braced, negated;
    switch (c)
      {
      /* A number of Perl escapes are not handled by PCRE. We give an explicit
      error. */
-     case 'l':
+     case CHAR_l:
-     case 'L':
+     case CHAR_L:
-     case 'N':
-     case 'u':
-     case 'U':
      *errorcodeptr = ERR37;
      break;
+     case CHAR_u:
+     if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       /* In JavaScript, \u must be followed by four hexadecimal numbers.
+       Otherwise it is a lowercase u letter. */
+       if (MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0
+         && MAX_255(ptr[2]) && (digitab[ptr[2]] & ctype_xdigit) != 0
+         && MAX_255(ptr[3]) && (digitab[ptr[3]] & ctype_xdigit) != 0
+         && MAX_255(ptr[4]) && (digitab[ptr[4]] & ctype_xdigit) != 0)
+         {
+         c = 0;
+         for (i = 0; i < 4; ++i)
+           {
+           register int 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
+           }
+         }
+       }
+     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)
+       {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
+     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
+       {
+       const pcre_uchar *p;
+       for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
+         if (*p != CHAR_MINUS && !IS_DIGIT(*p)) break;
+       if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
+         {
+         c = -ESC_k;
+         break;
+         }
+       braced = TRUE;
+       ptr++;
+       }
+     else braced = FALSE;
+     if (ptr[1] == CHAR_MINUS)
+       {
+       negated = TRUE;
+       ptr++;
+       }
+     else negated = FALSE;
+     /* The integer range is limited by the machine's int representation. */
+     c = 0;
+     while (IS_DIGIT(ptr[1]))
+       {
+       if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
+         {
+         c = -1;
+         break;
+         }
+       c = c * 10 + *(++ptr) - CHAR_0;
+       }
+     if (((unsigned int)c) > INT_MAX) /* Integer overflow */
+       {
+       while (IS_DIGIT(ptr[1]))
+         ptr++;
+       *errorcodeptr = ERR61;
+       break;
+       }
+     if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
+       {
+       *errorcodeptr = ERR57;
+       break;
+       }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
+       }
+     if (negated)
+       {
+       if (c > bracount)
+         {
+         *errorcodeptr = ERR15;
+         break;
+         }
+       c = bracount - (c - 1);
+       }
+     c = -(ESC_REF + c);
+     break;
      /* The handling of escape sequences consisting of a string of digits
      starting with one that is not zero is not straightforward. By experiment,
      the way Perl works seems to be as follows:
-Line 422 
 else
+Line 945 
 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)
+       c -= CHAR_0;
-         c = c * 10 + *(++ptr) - '0';
+       while (IS_DIGIT(ptr[1]))
+         {
+         if (((unsigned int)c) > INT_MAX / 10) /* Integer overflow */
+           {
+           c = -1;
+           break;
+           }
+         c = c * 10 + *(++ptr) - CHAR_0;
+         }
+       if (((unsigned int)c) > INT_MAX) /* Integer overflow */
+         {
+         while (IS_DIGIT(ptr[1]))
+           ptr++;
+         *errorcodeptr = ERR61;
+         break;
+         }
        if (c < 10 || c <= bracount)
          {
          c = -(ESC_REF + c);
-Line 443 
 else
+Line 981 
 else
      generates a binary zero byte and treats the digit as a following literal.
      Thus we have to pull back the pointer by one. */
-     if ((c = *ptr) >= '8')
+     if ((c = *ptr) >= CHAR_8)
        {
        ptr--;
        c = 0;
-Line 451 
 else
+Line 989 
 else
        }
      /* \0 always starts an octal number, but we may drop through to here with a
-     larger first octal digit. */
+     larger first octal digit. The original code used just to take the least
+     significant 8 bits of octal numbers (I think this is what early Perls used
-     case '0':
+     to do). Nowadays we allow for larger numbers in UTF-8 mode and 16-bit mode,
-     c -= '0';
+     but no more than 3 octal digits. */
-     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
-         c = c * 8 + *(++ptr) - '0';
+     case CHAR_0:
-     c &= 255;     /* Take least significant 8 bits */
+     c -= CHAR_0;
+     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
+         c = c * 8 + *(++ptr) - CHAR_0;
+ #ifdef COMPILE_PCRE8
+     if (!utf && c > 0xff) *errorcodeptr = ERR51;
+ #endif
      break;
-     /* \x is complicated when UTF-8 is enabled. \x{ddd} is a character number
+     /* \x is complicated. \x{ddd} is a character number which can be greater
-     which can be greater than 0xff, but only if the ddd are hex digits. */
+     than 0xff in utf or non-8bit mode, but only if the ddd are hex digits.
+     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 int 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)
- #ifdef SUPPORT_UTF8
-     if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)
        {
-       const uschar *pt = ptr + 2;
+       const pcre_uchar *pt = ptr + 2;
-       register int count = 0;
        c = 0;
-       while ((digitab[*pt] & ctype_xdigit) != 0)
+       while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0)
          {
-         int cc = *pt++;
+         register int cc = *pt++;
-         count++;
+         if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
- #if !EBCDIC    /* ASCII coding */
-         if (cc >= 'a') cc -= 32;               /* Convert to upper case */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-         c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+         if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
- #else          /* EBCDIC coding */
+         c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
-         if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */
+ #else           /* EBCDIC coding */
-         c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+         if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
+         c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
+ #endif
+ #ifdef COMPILE_PCRE8
+         if (c > (utf ? 0x10ffff : 0xff)) { c = -1; break; }
+ #else
+ #ifdef COMPILE_PCRE16
+         if (c > (utf ? 0x10ffff : 0xffff)) { c = -1; break; }
  #endif
+ #endif
+         }
+       if (c < 0)
+         {
+         while (MAX_255(*pt) && (digitab[*pt] & ctype_xdigit) != 0) pt++;
+         *errorcodeptr = ERR34;
          }
-       if (*pt == '}')
+       if (*pt == CHAR_RIGHT_CURLY_BRACKET)
          {
-         if (c < 0 || count > 8) *errorcodeptr = ERR34;
+         if (utf && c >= 0xd800 && c <= 0xdfff) *errorcodeptr = ERR73;
          ptr = pt;
          break;
          }
        /* If the sequence of hex digits does not end with '}', then we don't
        recognize this construct; fall through to the normal \x handling. */
        }
- #endif
-     /* Read just a single hex char */
+     /* Read just a single-byte hex-defined char */
      c = 0;
-     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
+     while (i++ < 2 && MAX_255(ptr[1]) && (digitab[ptr[1]] & ctype_xdigit) != 0)
        {
-       int cc;                               /* Some compilers don't like ++ */
+       int cc;                                  /* Some compilers don't like */
-       cc = *(++ptr);                        /* in initializers */
+       cc = *(++ptr);                           /* ++ in initializers */
- #if !EBCDIC    /* ASCII coding */
+ #ifndef EBCDIC  /* ASCII/UTF-8 coding */
-       if (cc >= 'a') cc -= 32;              /* Convert to upper case */
+       if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
- #else          /* EBCDIC coding */
+ #else           /* EBCDIC coding */
-       if (cc <= 'z') cc += 64;              /* Convert to upper case */
+       if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
-       c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
+       c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
  #endif
        }
      break;
-     /* Other special escapes not starting with a digit are straightforward */
+     /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
+     An error is given if the byte following \c is not an ASCII character. This
+     coding is ASCII-specific, but then the whole concept of \cx is
+     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
-     case 'c':
+     case CHAR_c:
      c = *(++ptr);
      if (c == 0)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
+ #ifndef EBCDIC    /* ASCII/UTF-8 coding */
-     /* A letter is upper-cased; then the 0x40 bit is flipped. This coding
+     if (c > 127)  /* Excludes all non-ASCII in either mode */
-     is ASCII-specific, but then the whole concept of \cx is ASCII-specific.
+       {
-     (However, an EBCDIC equivalent has now been added.) */
+       *errorcodeptr = ERR68;
+       break;
- #if !EBCDIC    /* ASCII coding */
+       }
-     if (c >= 'a' && c <= 'z') c -= 32;
+     if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
- #else          /* EBCDIC coding */
+ #else             /* EBCDIC coding */
-     if (c >= 'a' && c <= 'z') c += 64;
+     if (c >= CHAR_a && c <= CHAR_z) c += 64;
      c ^= 0xC0;
  #endif
      break;
      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
-     other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,
+     other alphanumeric following \ is an error if PCRE_EXTRA was set;
-     for Perl compatibility, it is a literal. This code looks a bit odd, but
+     otherwise, for Perl compatibility, it is a literal. This code looks a bit
-     there used to be some cases other than the default, and there may be again
+     odd, but there used to be some cases other than the default, and there may
-     in future, so I haven't "optimized" it. */
+     be again in future, so I haven't "optimized" it. */
      default:
      if ((options & PCRE_EXTRA) != 0) switch(c)
-Line 550 
 else
+Line 1135 
 else
      }
    }
+ /* Perl supports \N{name} for character names, as well as plain \N for "not
+ newline". PCRE does not support \N{name}. However, it does support
+ quantification such as \N{2,3}. */
+ if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
+      !is_counted_repeat(ptr+2))
+   *errorcodeptr = ERR37;
+ /* If PCRE_UCP is set, we change the values for \d etc. */
+ if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
+   c -= (ESC_DU - ESC_D);
+ /* Set the pointer to the final character before returning. */
  *ptrptr = ptr;
  return c;
  }
-Line 569 
 escape sequence.
+Line 1169 
 escape sequence.
  Argument:
    ptrptr         points to the pattern position pointer
    negptr         points to a boolean that is set TRUE for negation else FALSE
+   dptr           points to an int that is set to the detailed property value
    errorcodeptr   points to the error code variable
- Returns:     value from ucp_type_table, or -1 for an invalid type
+ Returns:         type value from ucp_type_table, or -1 for an invalid type
  */
  static int
- get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)
+ get_ucp(const pcre_uchar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
  {
  int c, i, bot, top;
- const uschar *ptr = *ptrptr;
+ const pcre_uchar *ptr = *ptrptr;
- char name[4];
+ pcre_uchar name[32];
  c = *(++ptr);
  if (c == 0) goto ERROR_RETURN;
  *negptr = FALSE;
- /* \P or \p can be followed by a one- or two-character name in {}, optionally
+ /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
- preceded by ^ for negation. */
+ negation. */
- if (c == '{')
+ if (c == CHAR_LEFT_CURLY_BRACKET)
    {
-   if (ptr[1] == '^')
+   if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
      {
      *negptr = TRUE;
      ptr++;
      }
-   for (i = 0; i <= 2; i++)
+   for (i = 0; i < (int)(sizeof(name) / sizeof(pcre_uchar)) - 1; i++)
      {
      c = *(++ptr);
      if (c == 0) goto ERROR_RETURN;
-     if (c == '}') break;
+     if (c == CHAR_RIGHT_CURLY_BRACKET) break;
      name[i] = c;
      }
-   if (c !='}')   /* Try to distinguish error cases */
+   if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
-     {
-     while (*(++ptr) != 0 && *ptr != '}');
-     if (*ptr == '}') goto UNKNOWN_RETURN; else goto ERROR_RETURN;
-     }
    name[i] = 0;
    }
-Line 624 
 else
+Line 1221 
 else
  /* Search for a recognized property name using binary chop */
  bot = 0;
- top = _pcre_utt_size;
+ top = PRIV(utt_size);
  while (bot < top)
    {
-   i = (bot + top)/2;
+   i = (bot + top) >> 1;
-   c = strcmp(name, _pcre_utt[i].name);
+   c = STRCMP_UC_C8(name, PRIV(utt_names) + PRIV(utt)[i].name_offset);
-   if (c == 0) return _pcre_utt[i].value;
+   if (c == 0)
+     {
+     *dptr = PRIV(utt)[i].value;
+     return PRIV(utt)[i].type;
+     }
    if (c > 0) bot = i + 1; else top = i;
    }
- UNKNOWN_RETURN:
  *errorcodeptr = ERR47;
  *ptrptr = ptr;
  return -1;
-Line 650 
 return -1;
+Line 1250 
 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 701 
 Returns:         pointer to '}' on succe
+Line 1268 
 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 710 
 int max = -1;
+Line 1277 
 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 + *p++ - CHAR_0;
  if (min < 0 || min > 65535)
    {
    *errorcodeptr = ERR5;
-Line 720 
 if (min < 0 || min > 65535)
+Line 1287 
 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 + *p++ - CHAR_0;
      if (max < 0 || max > 65535)
        {
        *errorcodeptr = ERR5;
-Line 750 
 return p;
+Line 1317 
 return p;
  /*************************************************
- *      Find first significant op code            *
+ *  Subroutine for finding forward reference      *
  *************************************************/
- /* This is called by several functions that scan a compiled expression looking
+ /* This recursive function is called only from find_parens() below. The
- for a fixed first character, or an anchoring op code etc. It skips over things
+ top-level call starts at the beginning of the pattern. All other calls must
- that do not influence this. For some calls, a change of option is important.
+ start at a parenthesis. It scans along a pattern's text looking for capturing
- For some calls, it makes sense to skip negative forward and all backward
+ subpatterns, and counting them. If it finds a named pattern that matches the
- assertions, and also the \b assertion; for others it does not.
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. Recursion is used to keep
+ track of subpatterns that reset the capturing group numbers - the (?| feature.
+ This function was originally called only from the second pass, in which we know
+ that if (?< or (?' or (?P< is encountered, the name will be correctly
+ terminated because that is checked in the first pass. There is now one call to
+ this function in the first pass, to check for a recursive back reference by
+ name (so that we can make the whole group atomic). In this case, we need check
+ only up to the current position in the pattern, and that is still OK because
+ and previous occurrences will have been checked. To make this work, the test
+ for "end of pattern" is a check against cd->end_pattern in the main loop,
+ instead of looking for a binary zero. This means that the special first-pass
+ call can adjust cd->end_pattern temporarily. (Checks for binary zero while
+ processing items within the loop are OK, because afterwards the main loop will
+ terminate.)
  Arguments:
-   code         pointer to the start of the group
+   ptrptr       address of the current character pointer (updated)
-   options      pointer to external options
+   cd           compile background data
-   optbit       the option bit whose changing is significant, or
+   name         name to seek, or NULL if seeking a numbered subpattern
-                  zero if none are
+   lorn         name length, or subpattern number if name is NULL
-   skipassert   TRUE if certain assertions are to be skipped
+   xmode        TRUE if we are in /x mode
+   utf          TRUE if we are in UTF-8 / UTF-16 mode
+   count        pointer to the current capturing subpattern number (updated)
- Returns:       pointer to the first significant opcode
+ Returns:       the number of the named subpattern, or -1 if not found
  */
- static const uschar*
+ static int
- first_significant_code(const uschar *code, int *options, int optbit,
+ find_parens_sub(pcre_uchar **ptrptr, compile_data *cd, const pcre_uchar *name, int lorn,
-   BOOL skipassert)
+   BOOL xmode, BOOL utf, int *count)
  {
- for (;;)
+ pcre_uchar *ptr = *ptrptr;
-   {
+ int start_count = *count;
-   switch ((int)*code)
+ int hwm_count = start_count;
-     {
+ BOOL dup_parens = FALSE;
-     case OP_OPT:
-     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
-       *options = (int)code[1];
-     code += 2;
-     break;
-     case OP_ASSERT_NOT:
+ /* If the first character is a parenthesis, check on the type of group we are
-     case OP_ASSERTBACK:
+ dealing with. The very first call may not start with a parenthesis. */
-     case OP_ASSERTBACK_NOT:
-     if (!skipassert) return code;
-     do code += GET(code, 1); while (*code == OP_ALT);
-     code += _pcre_OP_lengths[*code];
-     break;
-     case OP_WORD_BOUNDARY:
+ if (ptr[0] == CHAR_LEFT_PARENTHESIS)
-     case OP_NOT_WORD_BOUNDARY:
+   {
-     if (!skipassert) return code;
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
-     /* Fall through */
-     case OP_CALLOUT:
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
-     case OP_CREF:
-     case OP_BRANUMBER:
-     code += _pcre_OP_lengths[*code];
-     break;
-     default:
+   /* Handle a normal, unnamed capturing parenthesis. */
-     return code;
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
      }
-   }
- /* Control never reaches here */
- }
+   /* All cases now have (? at the start. Remember when we are in a group
+   where the parenthesis numbers are duplicated. */
+   else if (ptr[2] == CHAR_VERTICAL_LINE)
+     {
+     ptr += 3;
+     dup_parens = TRUE;
+     }
+   /* Handle comments; all characters are allowed until a ket is reached. */
- /*************************************************
+   else if (ptr[2] == CHAR_NUMBER_SIGN)
- *        Find the fixed length of a pattern      *
+     {
- *************************************************/
+     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
+     goto FAIL_EXIT;
+     }
- /* Scan a pattern and compute the fixed length of subject that will match it,
+   /* Handle a condition. If it is an assertion, just carry on so that it
- if the length is fixed. This is needed for dealing with backward assertions.
+   is processed as normal. If not, skip to the closing parenthesis of the
- In UTF8 mode, the result is in characters rather than bytes.
+   condition (there can't be any nested parens). */
- Arguments:
+   else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
-   code     points to the start of the pattern (the bracket)
+     {
-   options  the compiling options
+     ptr += 2;
+     if (ptr[1] != CHAR_QUESTION_MARK)
+       {
+       while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
+       if (*ptr != 0) ptr++;
+       }
+     }
- Returns:   the fixed length, or -1 if there is no fixed length,
+   /* Start with (? but not a condition. */
-              or -2 if \C was encountered
- */
+   else
+     {
+     ptr += 2;
+     if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
+     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
+     if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
+         ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
+       {
+       int term;
+       const 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 == ptr - thisname &&
+           STRNCMP_UC_UC(name, thisname, 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 == CHAR_BACKSLASH)
+     {
+     if (*(++ptr) == 0) goto FAIL_EXIT;
+     if (*ptr == CHAR_Q) for (;;)
+       {
+       while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
+       if (*ptr == 0) goto FAIL_EXIT;
+       if (*(++ptr) == CHAR_E) break;
+       }
+     continue;
+     }
+   /* Skip over character classes; this logic must be similar to the way they
+   are handled for real. If the first character is '^', skip it. Also, if the
+   first few characters (either before or after ^) are \Q\E or \E we skip them
+   too. This makes for compatibility with Perl. Note the use of STR macros to
+   encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
+   if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
+     {
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       if (ptr[1] == CHAR_BACKSLASH)
+         {
+         if (ptr[2] == CHAR_E)
+           ptr+= 2;
+         else if (STRNCMP_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 == 0) return -1;
+       if (*ptr == CHAR_BACKSLASH)
+         {
+         if (*(++ptr) == 0) goto FAIL_EXIT;
+         if (*ptr == CHAR_Q) for (;;)
+           {
+           while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
+           if (*ptr == 0) goto FAIL_EXIT;
+           if (*(++ptr) == CHAR_E) break;
+           }
+         continue;
+         }
+       }
+     continue;
+     }
+   /* Skip comments in /x mode */
+   if (xmode && *ptr == CHAR_NUMBER_SIGN)
+     {
+     ptr++;
+     while (*ptr != 0)
+       {
+       if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+       ptr++;
+ #ifdef SUPPORT_UTF
+       if (utf) FORWARDCHAR(ptr);
+ #endif
+       }
+     if (*ptr == 0) goto FAIL_EXIT;
+     continue;
+     }
+   /* Check for the special metacharacters */
+   if (*ptr == CHAR_LEFT_PARENTHESIS)
+     {
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, count);
+     if (rc > 0) return rc;
+     if (*ptr == 0) goto FAIL_EXIT;
+     }
+   else if (*ptr == CHAR_RIGHT_PARENTHESIS)
+     {
+     if (dup_parens && *count < hwm_count) *count = hwm_count;
+     goto FAIL_EXIT;
+     }
+   else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
+     {
+     if (*count > hwm_count) hwm_count = *count;
+     *count = start_count;
+     }
+   }
+ FAIL_EXIT:
+ *ptrptr = ptr;
+ return -1;
+ }
+ /*************************************************
+ *       Find forward referenced subpattern       *
+ *************************************************/
+ /* This function scans along a pattern's text looking for capturing
+ subpatterns, and counting them. If it finds a named pattern that matches the
+ name it is given, it returns its number. Alternatively, if the name is NULL, it
+ returns when it reaches a given numbered subpattern. This is used for forward
+ references to subpatterns. We used to be able to start this scan from the
+ current compiling point, using the current count value from cd->bracount, and
+ do it all in a single loop, but the addition of the possibility of duplicate
+ subpattern numbers means that we have to scan from the very start, in order to
+ take account of such duplicates, and to use a recursive function to keep track
+ of the different types of group.
+ Arguments:
+   cd           compile background data
+   name         name to seek, or NULL if seeking a numbered subpattern
+   lorn         name length, or subpattern number if name is NULL
+   xmode        TRUE if we are in /x mode
+   utf          TRUE if we are in UTF-8 / UTF-16 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. */
+ for (;;)
+   {
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf, &count);
+   if (rc > 0 || *ptr++ == 0) break;
+   }
+ return rc;
+ }
+ /*************************************************
+ *      Find first significant op code            *
+ *************************************************/
+ /* This is called by several functions that scan a compiled expression looking
+ for a fixed first character, or an anchoring op code etc. It skips over things
+ that do not influence this. For some calls, it makes sense to skip negative
+ forward and all backward assertions, and also the \b assertion; for others it
+ does not.
+ Arguments:
+   code         pointer to the start of the group
+   skipassert   TRUE if certain assertions are to be skipped
+ Returns:       pointer to the first significant opcode
+ */
+ static const pcre_uchar*
+ first_significant_code(const pcre_uchar *code, BOOL skipassert)
+ {
+ for (;;)
+   {
+   switch ((int)*code)
+     {
+     case OP_ASSERT_NOT:
+     case OP_ASSERTBACK:
+     case OP_ASSERTBACK_NOT:
+     if (!skipassert) return code;
+     do code += GET(code, 1); while (*code == OP_ALT);
+     code += PRIV(OP_lengths)[*code];
+     break;
+     case OP_WORD_BOUNDARY:
+     case OP_NOT_WORD_BOUNDARY:
+     if (!skipassert) return code;
+     /* Fall through */
+     case OP_CALLOUT:
+     case OP_CREF:
+     case OP_NCREF:
+     case OP_RREF:
+     case OP_NRREF:
+     case OP_DEF:
+     code += PRIV(OP_lengths)[*code];
+     break;
+     default:
+     return code;
+     }
+   }
+ /* Control never reaches here */
+ }
+ /*************************************************
+ *        Find the fixed length of a branch       *
+ *************************************************/
+ /* Scan a branch and compute the fixed length of subject that will match it,
+ if the length is fixed. This is needed for dealing with backward assertions.
+ In UTF8 mode, the result is in characters rather than bytes. The branch is
+ temporarily terminated with OP_END when this function is called.
+ This function is called when a backward assertion is encountered, so that if it
+ fails, the error message can point to the correct place in the pattern.
+ However, we cannot do this when the assertion contains subroutine calls,
+ because they can be forward references. We solve this by remembering this case
+ and doing the check at the end; a flag specifies which mode we are running in.
+ Arguments:
+   code     points to the start of the pattern (the bracket)
+   utf      TRUE in UTF-8 / UTF-16 mode
+   atend    TRUE if called when the pattern is complete
+   cd       the "compile data" structure
+ Returns:   the fixed length,
+              or -1 if there is no fixed length,
+              or -2 if \C was encountered (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 842 
 branch, check the length against that of
+Line 1709 
 branch, check the length against that of
  for (;;)
    {
    int d;
+   pcre_uchar *ce, *cs;
    register int op = *cc;
-   if (op >= OP_BRA) op = OP_BRA;
    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, 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 873 
 for (;;)
+Line 1749 
 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 880 
 for (;;)
+Line 1771 
 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_BRANUMBER:
+     case OP_PRUNE_ARG:
-     case OP_CREF:
+     case OP_SKIP_ARG:
-     case OP_OPT:
+     case OP_THEN_ARG:
+     cc += cc[1] + PRIV(OP_lengths)[*cc];
+     break;
      case OP_CALLOUT:
-     case OP_SOD:
-     case OP_SOM:
-     case OP_EOD:
-     case OP_EODN:
      case OP_CIRC:
+     case OP_CIRCM:
+     case OP_CLOSE:
+     case OP_COMMIT:
+     case OP_CREF:
+     case OP_DEF:
      case OP_DOLL:
+     case OP_DOLLM:
+     case OP_EOD:
+     case OP_EODN:
+     case OP_FAIL:
+     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 918 
 for (;;)
+Line 1827 
 for (;;)
      need to skip over a multibyte character in UTF8 mode.  */
      case OP_EXACT:
+     case OP_EXACTI:
+     case OP_NOTEXACT:
+     case OP_NOTEXACTI:
      branchlength += GET2(cc,1);
-     cc += 4;
+     cc += 2 + IMM2_SIZE;
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
-     if ((options & PCRE_UTF8) != 0)
+     if (utf && HAS_EXTRALEN(cc[-1])) cc += GET_EXTRALEN(cc[-1]);
-       {
-       while((*cc & 0x80) == 0x80) cc++;
-       }
  #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 */
      case OP_PROP:
      case OP_NOTPROP:
-     cc++;
+     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 947 
 for (;;)
+Line 1861 
 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
+ #if defined SUPPORT_UTF || defined COMPILE_PCRE16
      case OP_XCLASS:
-     cc += GET(cc, 1) - 33;
+     cc += GET(cc, 1) - PRIV(OP_lengths)[OP_CLASS];
      /* Fall through */
  #endif
      case OP_CLASS:
      case OP_NCLASS:
-     cc += 33;
+     cc += PRIV(OP_lengths)[OP_CLASS];
      switch (*cc)
        {
+       case OP_CRPLUS:
+       case OP_CRMINPLUS:
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRQUERY:
-Line 978 
 for (;;)
+Line 1896 
 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);
-       cc += 5;
+       cc += 1 + 2 * IMM2_SIZE;
        break;
        default:
-Line 990 
 for (;;)
+Line 1908 
 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 1001 
 for (;;)
+Line 2002 
 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 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)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c > OP_BRA)
+   /* XCLASS is used for classes that cannot be represented just by a bit
+   map. This includes negated single high-valued characters. The length in
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Handle recursion */
+   else if (c == OP_REVERSE)
+     {
+     if (number < 0) return (pcre_uchar *)code;
+     code += PRIV(OP_lengths)[c];
+     }
+   /* Handle capturing bracket */
+   else if (c == OP_CBRA || c == OP_SCBRA ||
+            c == OP_CBRAPOS || c == OP_SCBRAPOS)
      {
-     int n = c - OP_BRA;
+     int n = GET2(code, 1+LINK_SIZE);
-     if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);
+     if (n == number) return (pcre_uchar *)code;
-     if (n == number) return (uschar *)code;
+     code += PRIV(OP_lengths)[c];
-     code += _pcre_OP_lengths[OP_BRA];
      }
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
- #ifdef SUPPORT_UTF8
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       if (code[1 + IMM2_SIZE] == OP_PROP
+         || code[1 + IMM2_SIZE] == OP_NOTPROP) code += 2;
+       break;
-     /* In UTF-8 mode, opcodes that are followed by a character may be followed
+       case OP_MARK:
-     by a multi-byte character. The length in the table is a minimum, so we have
+       case OP_PRUNE_ARG:
-     to scan along to skip the extra bytes. All opcodes are less than 128, so we
+       case OP_SKIP_ARG:
-     can use relatively efficient code. */
+       code += code[1];
+       break;
+       case OP_THEN_ARG:
+       code += code[1];
+       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 (utf8) switch(c)
+ #ifdef SUPPORT_UTF
+     if (utf) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_MINQUERYI:
-       break;
+       case OP_POSQUERY:
+       case OP_POSQUERYI:
-       /* XCLASS is used for classes that cannot be represented just by a bit
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
+ #else
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1084 
 instance of OP_RECURSE.
+Line 2152 
 instance of OP_RECURSE.
  Arguments:
    code        points to start of expression
-   utf8        TRUE in UTF-8 mode
+   utf         TRUE in UTF-8 / UTF-16 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)
  {
- #ifndef SUPPORT_UTF8
- utf8 = utf8;               /* Stop pedantic compilers complaining */
- #endif
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
-   else if (c == OP_RECURSE) return code;
+   if (c == OP_RECURSE) return code;
-   else if (c > OP_BRA)
-     {
+   /* XCLASS is used for classes that cannot be represented just by a bit
-     code += _pcre_OP_lengths[OP_BRA];
+   map. This includes negated single high-valued characters. The length in
-     }
+   the table is zero; the actual length is stored in the compiled code. */
+   if (c == OP_XCLASS) code += GET(code, 1);
+   /* Otherwise, we can get the item's length from the table, except that for
+   repeated character types, we have to test for \p and \P, which have an extra
+   two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
+   must add in its length. */
    else
      {
-     code += _pcre_OP_lengths[c];
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
+       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;
- #ifdef SUPPORT_UTF8
+       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 scan along to skip the extra bytes. All opcodes are less than 128, so we
+     to arrange to skip the extra bytes. */
-     can use relatively efficient code. */
-     if (utf8) switch(c)
+ #ifdef SUPPORT_UTF
+     if (utf) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
+       case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
+       case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
+       case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
-       while ((*code & 0xc0) == 0x80) code++;
+       case OP_MINQUERYI:
-       break;
+       case OP_POSQUERY:
+       case OP_POSQUERYI:
-       /* XCLASS is used for classes that cannot be represented just by a bit
+       if (HAS_EXTRALEN(code[-1])) code += GET_EXTRALEN(code[-1]);
-       map. This includes negated single high-valued characters. The length in
-       the table is zero; the actual length is stored in the compiled code. */
-       case OP_XCLASS:
-       code += GET(code, 1) + 1;
        break;
        }
+ #else
+     (void)(utf);  /* Keep compiler happy by referencing function argument */
  #endif
      }
    }
-Line 1152 
 for (;;)
+Line 2268 
 for (;;)
  *************************************************/
  /* This function scans through a branch of a compiled pattern to see whether it
- can match the empty string or not. It is called only from could_be_empty()
+ can match the empty string or not. It is called from could_be_empty()
- below. Note that first_significant_code() skips over assertions. If we hit an
+ below and from compile_branch() when checking for an unlimited repeat of a
- unclosed bracket, we return "empty" - this means we've struck an inner bracket
+ group that can match nothing. Note that first_significant_code() skips over
- whose current branch will already have been scanned.
+ backward and negative forward assertions when its final argument is TRUE. If we
+ hit an unclosed bracket, we return "empty" - this means we've struck an inner
+ bracket whose current branch will already have been scanned.
  Arguments:
    code        points to start of search
    endcode     points to where to stop
-   utf8        TRUE if in UTF8 mode
+   utf         TRUE if in UTF-8 / UTF-16 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;
- for (code = first_significant_code(code + 1 + LINK_SIZE, 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;
-   if (c >= OP_BRA)
+   /* Skip over forward assertions; the other assertions are skipped by
+   first_significant_code() with a TRUE final argument. */
+   if (c == OP_ASSERT)
      {
-     BOOL empty_branch;
+     do code += GET(code, 1); while (*code == OP_ALT);
-     if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
+     c = *code;
+     continue;
+     }
-     /* Scan a closed bracket */
+   /* For a recursion/subroutine call, if its end has been reached, which
+   implies a backward reference subroutine call, we can scan it. If it's a
+   forward reference subroutine call, we can't. To detect forward reference
+   we have to scan up the list that is kept in the workspace. This function is
+   called only when doing the real compile, not during the pre-compile that
+   measures the size of the compiled pattern. */
+   if (c == OP_RECURSE)
+     {
+     const pcre_uchar *scode;
+     BOOL empty_branch;
+     /* Test for forward reference */
+     for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
+       if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
+     /* Not a forward reference, test for completed backward reference */
+     empty_branch = FALSE;
+     scode = cd->start_code + GET(code, 1);
+     if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
+     /* Completed backwards reference */
-     empty_branch = FALSE;
      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 */
-     code += 1 + LINK_SIZE;
      c = *code;
+     continue;
      }
-   else switch (c)
+   /* Handle the other opcodes */
+   switch (c)
      {
-     /* Check for quantifiers after a class */
+     /* Check for quantifiers after a class. XCLASS is used for classes that
+     cannot be represented just by a bit map. This includes negated single
+     high-valued characters. The length in 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 1247 
 for (code = first_significant_code(code
+Line 2458 
 for (code = first_significant_code(code
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      case OP_PLUS:
      case OP_MINPLUS:
+     case OP_POSPLUS:
      case OP_EXACT:
      case OP_NOTPLUS:
      case OP_NOTMINPLUS:
+     case OP_NOTPOSPLUS:
      case OP_NOTEXACT:
      case OP_TYPEPLUS:
      case OP_TYPEMINPLUS:
+     case OP_TYPEPOSPLUS:
      case OP_TYPEEXACT:
      return FALSE;
+     /* These are going to continue, as they may be empty, but we have to
+     fudge the length for the \p and \P cases. */
+     case OP_TYPESTAR:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEQUERY:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEPOSQUERY:
+     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+     break;
+     /* Same for these */
+     case OP_TYPEUPTO:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPOSUPTO:
+     if (code[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, QUERY, MINQUERY, UPTO, and MINUPTO  may be
+     /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
-     followed by a multibyte character */
+     MINUPTO, and POSUPTO may be followed by a multibyte character */
- #ifdef SUPPORT_UTF8
+ #ifdef SUPPORT_UTF
      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:
-     if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
+     case OP_MINUPTOI:
+     case OP_POSUPTO:
+     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 1299 
 return TRUE;
+Line 2570 
 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 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 1328 
 return TRUE;
+Line 2603 
 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 */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
- if (*(++ptr) == '^') ptr++;
+ for (++ptr; *ptr != 0; ptr++)
- while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
- if (*ptr == terminator && ptr[1] == ']')
    {
-   *endptr = ptr;
+   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
-   return TRUE;
+     ptr++;
+   else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+   else
+     {
+     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
+          (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
+           ptr[1] == CHAR_EQUALS_SIGN) &&
+         check_posix_syntax(ptr, endptr))
+       return FALSE;
+     }
    }
  return FALSE;
  }
-Line 1373 
 Returns:     a value representing the na
+Line 2685 
 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, len) == 0) return yield;
+   pn += posix_name_lengths[yield] + 1;
    yield++;
    }
  return -1;
-Line 1394 
 return -1;
+Line 2708 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That is the job of this function. Before it is called, the
+ have their offsets adjusted. That one of the jobs of this function. Before it
- partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
+ This function has been extended with the possibility of forward references for
+ recursions and subroutine calls. It must also check the list of such references
+ for the group we are dealing with. If it finds that one of the recursions in
+ the current group is on this list, it adjusts the offset in the list, not the
+ value in the reference (which is a group number).
  Arguments:
    group      points to the start of the group
    adjust     the amount by which the group is to be moved
-   utf8       TRUE in UTF-8 mode
+   utf        TRUE in UTF-8 / UTF-16 mode
    cd         contains pointers to tables etc.
+   save_hwm   the hwm forward reference pointer at the start of the group
  Returns:     nothing
  */
  static void
- adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)
+ adjust_recurse(pcre_uchar *group, int adjust, BOOL utf, compile_data *cd,
+   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 = GET(ptr, 1);
+   int offset;
-   if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+   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);
+     if (cd->start_code + offset == ptr + 1)
+       {
+       PUT(hc, 0, offset + adjust);
+       break;
+       }
+     }
+   /* Otherwise, adjust the recursion offset if it's after the start of this
+   group. */
+   if (hc >= cd->hwm)
+     {
+     offset = GET(ptr, 1);
+     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
+     }
    ptr += 1 + LINK_SIZE;
    }
  }
-Line 1438 
 Arguments:
+Line 2785 
 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 1467 
 Returns:             nothing
+Line 2814 
 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 1495 
 Yield:        TRUE when range returned;
+Line 2842 
 Yield:        TRUE when range returned;
  */
  static BOOL
- get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)
+ get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
+   unsigned int *odptr)
  {
- int c, chartype, othercase, next;
+ unsigned int c, othercase, next;
  for (c = *cptr; c <= d; c++)
-   {
+   { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
-   if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)
-     break;
-   }
  if (c > d) return FALSE;
-Line 1512 
 next = othercase + 1;
+Line 2857 
 next = othercase + 1;
  for (++c; c <= d; c++)
    {
-   if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||
+   if (UCD_OTHERCASE(c) != next) break;
-         othercase != next)
-     break;
    next++;
    }
-Line 1523 
 for (++c; c <= d; c++)
+Line 2866 
 for (++c; c <= d; c++)
  return TRUE;
  }
+ /*************************************************
+ *        Check a character and a property        *
+ *************************************************/
+ /* This function is called by check_auto_possessive() when a property item
+ is adjacent to a fixed character.
+ Arguments:
+   c            the character
+   ptype        the property type
+   pdata        the data for the type
+   negated      TRUE if it's a negated property (\P or \p{^)
+ Returns:       TRUE if auto-possessifying is OK
+ */
+ static BOOL
+ check_char_prop(int c, int ptype, int pdata, BOOL negated)
+ {
+ const ucd_record *prop = GET_UCD(c);
+ switch(ptype)
+   {
+   case PT_LAMP:
+   return (prop->chartype == ucp_Lu ||
+           prop->chartype == ucp_Ll ||
+           prop->chartype == ucp_Lt) == negated;
+   case PT_GC:
+   return (pdata == 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;
+   }
+ return FALSE;
+ }
  #endif  /* SUPPORT_UCP */
+ /*************************************************
+ *     Check if auto-possessifying is possible    *
+ *************************************************/
+ /* This function is called for unlimited repeats of certain items, to see
+ whether the next thing could possibly match the repeated item. If not, it makes
+ sense to automatically possessify the repeated item.
+ Arguments:
+   previous      pointer to the repeated opcode
+   utf           TRUE in UTF-8 / UTF-16 mode
+   ptr           next character in pattern
+   options       options bits
+   cd            contains pointers to tables etc.
+ Returns:        TRUE if possessifying is wanted
+ */
+ static BOOL
+ check_auto_possessive(const pcre_uchar *previous, BOOL utf,
+   const pcre_uchar *ptr, int options, compile_data *cd)
+ {
+ pcre_int32 c, next;
+ int op_code = *previous++;
+ /* Skip whitespace and comments in extended mode */
+ if ((options & PCRE_EXTENDED) != 0)
+   {
+   for (;;)
+     {
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     if (*ptr == CHAR_NUMBER_SIGN)
+       {
+       ptr++;
+       while (*ptr != 0)
+         {
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
+       }
+     else break;
+     }
+   }
+ /* If the next item is one that we can handle, get its value. A non-negative
+ value is a character, a negative value is an escape value. */
+ if (*ptr == CHAR_BACKSLASH)
+   {
+   int temperrorcode = 0;
+   next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
+   if (temperrorcode != 0) return FALSE;
+   ptr++;    /* Point after the escape sequence */
+   }
+ else if (!MAX_255(*ptr) || (cd->ctypes[*ptr] & ctype_meta) == 0)
+   {
+ #ifdef SUPPORT_UTF
+   if (utf) { GETCHARINC(next, ptr); } else
+ #endif
+   next = *ptr++;
+   }
+ else return FALSE;
+ /* Skip whitespace and comments in extended mode */
+ if ((options & PCRE_EXTENDED) != 0)
+   {
+   for (;;)
+     {
+     while (MAX_255(*ptr) && (cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
+     if (*ptr == CHAR_NUMBER_SIGN)
+       {
+       ptr++;
+       while (*ptr != 0)
+         {
+         if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF
+         if (utf) FORWARDCHAR(ptr);
+ #endif
+         }
+       }
+     else break;
+     }
+   }
+ /* If the next thing is itself optional, we have to give up. */
+ if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+   STRNCMP_UC_C8(ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+     return FALSE;
+ /* Now compare the next item with the previous opcode. First, handle cases when
+ the next item is a character. */
+ if (next >= 0) switch(op_code)
+   {
+   case OP_CHAR:
+ #ifdef SUPPORT_UTF
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   return c != next;
+   /* For CHARI (caseless character) we must check the other case. If we have
+   Unicode property support, we can use it to test the other case of
+   high-valued characters. */
+   case OP_CHARI:
+ #ifdef SUPPORT_UTF
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   if (c == next) return FALSE;
+ #ifdef SUPPORT_UTF
+   if (utf)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE((unsigned int)next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)c != othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF */
+   return (c != TABLE_GET((unsigned int)next, cd->fcc, next));  /* Non-UTF-8 mode */
+   /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
+   opcodes are not used for multi-byte characters, because they are coded using
+   an XCLASS instead. */
+   case OP_NOT:
+   return (c = *previous) == next;
+   case OP_NOTI:
+   if ((c = *previous) == next) return TRUE;
+ #ifdef SUPPORT_UTF
+   if (utf)
+     {
+     unsigned int othercase;
+     if (next < 128) othercase = cd->fcc[next]; else
+ #ifdef SUPPORT_UCP
+     othercase = UCD_OTHERCASE(next);
+ #else
+     othercase = NOTACHAR;
+ #endif
+     return (unsigned int)c == othercase;
+     }
+   else
+ #endif  /* SUPPORT_UTF */
+   return (c == (int)(TABLE_GET((unsigned int)next, cd->fcc, next)));  /* Non-UTF-8 mode */
+   /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
+   When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
+   case OP_DIGIT:
+   return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
+   case OP_NOT_DIGIT:
+   return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
+   case OP_WHITESPACE:
+   return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
+   case OP_NOT_WHITESPACE:
+   return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
+   case OP_WORDCHAR:
+   return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
+   case OP_NOT_WORDCHAR:
+   return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
+   case OP_HSPACE:
+   case OP_NOT_HSPACE:
+   switch(next)
+     {
+     case 0x09:
+     case 0x20:
+     case 0xa0:
+     case 0x1680:
+     case 0x180e:
+     case 0x2000:
+     case 0x2001:
+     case 0x2002:
+     case 0x2003:
+     case 0x2004:
+     case 0x2005:
+     case 0x2006:
+     case 0x2007:
+     case 0x2008:
+     case 0x2009:
+     case 0x200A:
+     case 0x202f:
+     case 0x205f:
+     case 0x3000:
+     return op_code == OP_NOT_HSPACE;
+     default:
+     return op_code != OP_NOT_HSPACE;
+     }
+   case OP_ANYNL:
+   case OP_VSPACE:
+   case OP_NOT_VSPACE:
+   switch(next)
+     {
+     case 0x0a:
+     case 0x0b:
+     case 0x0c:
+     case 0x0d:
+     case 0x85:
+     case 0x2028:
+     case 0x2029:
+     return op_code == OP_NOT_VSPACE;
+     default:
+     return op_code != OP_NOT_VSPACE;
+     }
+ #ifdef SUPPORT_UCP
+   case OP_PROP:
+   return check_char_prop(next, previous[0], previous[1], FALSE);
+   case OP_NOTPROP:
+   return check_char_prop(next, previous[0], previous[1], TRUE);
+ #endif
+   default:
+   return FALSE;
+   }
+ /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
+ is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
+ generated only when PCRE_UCP is *not* set, that is, when only ASCII
+ characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
+ replaced by OP_PROP codes when PCRE_UCP is set. */
+ switch(op_code)
+   {
+   case OP_CHAR:
+   case OP_CHARI:
+ #ifdef SUPPORT_UTF
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
+ #endif
+   switch(-next)
+     {
+     case ESC_d:
+     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
+     case ESC_D:
+     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
+     case ESC_s:
+     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
+     case ESC_S:
+     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
+     case ESC_w:
+     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
+     case ESC_W:
+     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
+     case ESC_h:
+     case ESC_H:
+     switch(c)
+       {
+       case 0x09:
+       case 0x20:
+       case 0xa0:
+       case 0x1680:
+       case 0x180e:
+       case 0x2000:
+       case 0x2001:
+       case 0x2002:
+       case 0x2003:
+       case 0x2004:
+       case 0x2005:
+       case 0x2006:
+       case 0x2007:
+       case 0x2008:
+       case 0x2009:
+       case 0x200A:
+       case 0x202f:
+       case 0x205f:
+       case 0x3000:
+       return -next != ESC_h;
+       default:
+       return -next == ESC_h;
+       }
+     case ESC_v:
+     case ESC_V:
+     switch(c)
+       {
+       case 0x0a:
+       case 0x0b:
+       case 0x0c:
+       case 0x0d:
+       case 0x85:
+       case 0x2028:
+       case 0x2029:
+       return -next != ESC_v;
+       default:
+       return -next == ESC_v;
+       }
+     /* When PCRE_UCP is set, these values get generated for \d etc. Find
+     their substitutions and process them. The result will always be either
+     -ESC_p or -ESC_P. Then fall through to process those values. */
+ #ifdef SUPPORT_UCP
+     case ESC_du:
+     case ESC_DU:
+     case ESC_wu:
+     case ESC_WU:
+     case ESC_su:
+     case ESC_SU:
+       {
+       int temperrorcode = 0;
+       ptr = substitutes[-next - ESC_DU];
+       next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
+       }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       int ptype, pdata, errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
+       if (ptype < 0) return FALSE;
+       ptr++;      /* Point past the final curly ket */
+       /* If the property item is optional, we have to give up. (When generated
+       from \d etc by PCRE_UCP, this test will have been applied much earlier,
+       to the original \d etc. At this point, ptr will point to a zero byte. */
+       if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+         STRNCMP_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, (next == -ESC_P) != negated);
+       }
+ #endif
+     default:
+     return FALSE;
+     }
+   /* In principle, support for Unicode properties should be integrated here as
+   well. It means re-organizing the above code so as to get hold of the property
+   values before switching on the op-code. However, I wonder how many patterns
+   combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
+   these op-codes are never generated.) */
+   case OP_DIGIT:
+   return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
+          next == -ESC_h || next == -ESC_v || next == -ESC_R;
+   case OP_NOT_DIGIT:
+   return next == -ESC_d;
+   case OP_WHITESPACE:
+   return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
+   case OP_NOT_WHITESPACE:
+   return next == -ESC_s || next == -ESC_h || next == -ESC_v;
+   case OP_HSPACE:
+   return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
+          next == -ESC_w || next == -ESC_v || next == -ESC_R;
+   case OP_NOT_HSPACE:
+   return next == -ESC_h;
+   /* Can't have \S in here because VT matches \S (Perl anomaly) */
+   case OP_ANYNL:
+   case OP_VSPACE:
+   return next == -ESC_V || next == -ESC_d || next == -ESC_w;
+   case OP_NOT_VSPACE:
+   return next == -ESC_v || next == -ESC_R;
+   case OP_WORDCHAR:
+   return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
+          next == -ESC_v || next == -ESC_R;
+   case OP_NOT_WORDCHAR:
+   return next == -ESC_w || next == -ESC_d;
+   default:
+   return FALSE;
+   }
+ /* Control does not reach here */
+ }
  /*************************************************
  *           Compile one branch                   *
  *************************************************/
- /* Scan the pattern, compiling it into the code vector. If the options are
+ /* Scan the pattern, compiling it into the a vector. If the options are
  changed during the branch, the pointer is used to change the external options
- bits.
+ bits. This function is used during the pre-compile phase when we are trying
+ to find out the amount of memory needed, as well as during the real compile
+ phase. The value of lengthptr distinguishes the two phases.
  Arguments:
    optionsptr     pointer to the option bits
-   brackets       points to number of extracting brackets used
    codeptr        points to the pointer to the current code point
    ptrptr         points to the current pattern pointer
    errorcodeptr   points to error code variable
-   firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
+   firstcharptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
-   reqbyteptr     set to the last literal character required, else < 0
+   reqcharptr     set to the last literal character required, else < 0
    bcptr          points to current branch chain
+   cond_depth     conditional nesting depth
    cd             contains pointers to tables etc.
+   lengthptr      NULL during the real compile phase
+                  points to length accumulator during pre-compile phase
  Returns:         TRUE on success
                   FALSE, with *errorcodeptr set non-zero on error
  */
  static BOOL
- compile_branch(int *optionsptr, int *brackets, uschar **codeptr,
+ compile_branch(int *optionsptr, pcre_uchar **codeptr,
-   const uschar **ptrptr, int *errorcodeptr, int *firstbyteptr,
+   const pcre_uchar **ptrptr, int *errorcodeptr, pcre_int32 *firstcharptr,
-   int *reqbyteptr, branch_chain *bcptr, compile_data *cd)
+   pcre_int32 *reqcharptr, 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_int32 firstchar, reqchar;
- int zeroreqbyte, zerofirstbyte;
+ pcre_int32 zeroreqchar, zerofirstchar;
- int req_caseopt, reqvary, tempreqvary;
+ pcre_int32 req_caseopt, reqvary, tempreqvary;
- int condcount = 0;
+ int options = *optionsptr;               /* May change dynamically */
- int options = *optionsptr;
  int after_manual_callout = 0;
+ int length_prevgroup = 0;
  register int c;
- register uschar *code = *codeptr;
+ register pcre_uchar *code = *codeptr;
- uschar *tempcode;
+ pcre_uchar *last_code = code;
+ 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 classbits[32];
+ pcre_uchar *previous_callout = NULL;
+ pcre_uchar *save_hwm = NULL;
- #ifdef SUPPORT_UTF8
+ pcre_uint8 classbits[32];
- BOOL class_utf8;
- BOOL utf8 = (options & PCRE_UTF8) != 0;
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
- uschar *class_utf8data;
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
- uschar utf8_char[6];
+ dynamically as we process the pattern. */
+ #ifdef SUPPORT_UTF
+ /* PCRE_UTF16 has the same value as PCRE_UTF8. */
+ BOOL utf = (options & PCRE_UTF8) != 0;
+ pcre_uchar utf_chars[6];
  #else
- BOOL utf8 = FALSE;
+ BOOL utf = FALSE;
+ #endif
+ /* Helper variables for OP_XCLASS opcode (for characters > 255). */
+ #if defined SUPPORT_UTF || !defined COMPILE_PCRE8
+ BOOL xclass;
+ pcre_uchar *class_uchardata;
+ pcre_uchar *class_uchardata_base;
+ #endif
+ #ifdef PCRE_DEBUG
+ if (lengthptr != NULL) DPRINTF((">> start branch\n"));
  #endif
  /* Set up the default and non-default settings for greediness */
-Line 1591 
 greedy_non_default = greedy_default ^ 1;
+Line 3440 
 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 = 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;
-   int class_charcount;
+   BOOL is_recurse;
-   int class_lastchar;
+   BOOL reset_bracount;
+   int class_has_8bitchar;
+   int class_single_char;
    int newoptions;
    int recno;
+   int refsign;
    int skipbytes;
-   int subreqbyte;
+   int subreqchar;
-   int subfirstbyte;
+   int subfirstchar;
+   int terminator;
    int mclength;
-   uschar mcbuffer[8];
+   int tempbracount;
+   pcre_uchar mcbuffer[8];
-   /* Next byte in the pattern */
+   /* Get next character in the pattern */
    c = *ptr;
-   /* If in \Q...\E, check for the end; if not, we have a literal */
+   /* If we are at the end of a nested substitution, revert to the outer level
+   string. Nesting only happens one level deep. */
-   if (inescq && c != 0)
+   if (c == 0 && nestptr != NULL)
      {
-     if (c == '\\' && ptr[1] == 'E')
+     ptr = nestptr;
-       {
+     nestptr = NULL;
-       inescq = FALSE;
+     c = *ptr;
+     }
+   /* If we are in the pre-compile phase, accumulate the length used for the
+   previous cycle of this loop. */
+   if (lengthptr != NULL)
+     {
+ #ifdef PCRE_DEBUG
+     if (code > cd->hwm) cd->hwm = code;                 /* High water info */
+ #endif
+     if (code > cd->start_workspace + cd->workspace_size -
+         WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
+       {
+       *errorcodeptr = ERR52;
+       goto FAILED;
+       }
+     /* There is at least one situation where code goes backwards: this is the
+     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
+     the class is simply eliminated. However, it is created first, so we have to
+     allow memory for it. Therefore, don't ever reduce the length at this point.
+     */
+     if (code < last_code) code = last_code;
+     /* Paranoid check for integer overflow */
+     if (OFLOW_MAX - *lengthptr < code - last_code)
+       {
+       *errorcodeptr = ERR20;
+       goto FAILED;
+       }
+     *lengthptr += (int)(code - last_code);
+     DPRINTF(("length=%d added %d c=%c (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,
+     if "previous" is NULL, reset the current code pointer to the start. */
+     if (previous != NULL)
+       {
+       if (previous > orig_code)
+         {
+         memmove(orig_code, previous, IN_UCHARS(code - previous));
+         code -= previous - orig_code;
+         previous = orig_code;
+         }
+       }
+     else code = orig_code;
+     /* Remember where this code item starts so we can pick up the length
+     next time round. */
+     last_code = code;
+     }
+   /* In the real compile phase, just check the workspace used by the forward
+   reference list. */
+   else if (cd->hwm > cd->start_workspace + cd->workspace_size -
+            WORK_SIZE_SAFETY_MARGIN)
+     {
+     *errorcodeptr = ERR52;
+     goto FAILED;
+     }
+   /* If in \Q...\E, check for the end; if not, we have a literal */
+   if (inescq && c != 0)
+     {
+     if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
+       {
+       inescq = FALSE;
        ptr++;
        continue;
        }
-Line 1643 
 for (;; ptr++)
+Line 3575 
 for (;; ptr++)
        {
        if (previous_callout != NULL)
          {
-         complete_callout(previous_callout, ptr, cd);
+         if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
+           complete_callout(previous_callout, ptr, cd);
          previous_callout = NULL;
          }
        if ((options & PCRE_AUTO_CALLOUT) != 0)
-Line 1658 
 for (;; ptr++)
+Line 3591 
 for (;; ptr++)
    /* Fill in length of a previous callout, except when the next thing is
    a quantifier. */
-   is_quantifier = c == '*' || c == '+' || c == '?' ||
+   is_quantifier =
-     (c == '{' && is_counted_repeat(ptr+1));
+     c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
+     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
    if (!is_quantifier && previous_callout != NULL &&
         after_manual_callout-- <= 0)
      {
-     complete_callout(previous_callout, ptr, cd);
+     if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
+       complete_callout(previous_callout, ptr, cd);
      previous_callout = NULL;
      }
-   /* In extended mode, skip white space and comments */
+   /* In extended mode, skip white space and comments. */
    if ((options & PCRE_EXTENDED) != 0)
      {
-     if ((cd->ctypes[c] & ctype_space) != 0) continue;
+     if (MAX_255(*ptr) && (cd->ctypes[c] & ctype_space) != 0) continue;
-     if (c == '#')
+     if (c == CHAR_NUMBER_SIGN)
        {
-       /* The space before the ; is to avoid a warning on a silly compiler
+       ptr++;
-       on the Macintosh. */
+       while (*ptr != 0)
-       while ((c = *(++ptr)) != 0 && c != NEWLINE) ;
+         {
        if (c