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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 210 by ph10,
Wed Aug  8 14:24:50 2007 UTC
+revision 345 by ph10,
Mon Apr 28 15:10:02 2008 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2008 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 43 
 supporting internal functions that are n
  #ifdef HAVE_CONFIG_H
- #include <config.h>
+ #include "config.h"
  #endif
  #define NLBLOCK cd             /* Block containing newline information */
 Line 140 
 static const short int escapes[] = {
  #endif
- /* Table of special "verbs" like (*PRUNE) */
+ /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
+ searched linearly. Put all the names into a single string, in order to reduce
+ the number of relocations when a shared library is dynamically linked. */
  typedef struct verbitem {
-   const char *name;
    int   len;
    int   op;
  } verbitem;
- static verbitem verbs[] = {
+ static const char verbnames[] =
-   { "ACCEPT", 6, OP_ACCEPT },
+   "ACCEPT\0"
-   { "COMMIT", 6, OP_COMMIT },
+   "COMMIT\0"
-   { "F",      1, OP_FAIL },
+   "F\0"
-   { "FAIL",   4, OP_FAIL },
+   "FAIL\0"
-   { "PRUNE",  5, OP_PRUNE },
+   "PRUNE\0"
-   { "SKIP",   4, OP_SKIP  },
+   "SKIP\0"
-   { "THEN",   4, OP_THEN  }
+   "THEN";
+ static const verbitem verbs[] = {
+   { 6, OP_ACCEPT },
+   { 6, OP_COMMIT },
+   { 1, OP_FAIL },
+   { 4, OP_FAIL },
+   { 5, OP_PRUNE },
+   { 4, OP_SKIP  },
+   { 4, OP_THEN  }
  };
- static int verbcount = sizeof(verbs)/sizeof(verbitem);
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
- /* Tables of names of POSIX character classes and their lengths. The list is
+ /* Tables of names of POSIX character classes and their lengths. The names are
- terminated by a zero length entry. The first three must be alpha, lower, upper,
+ now all in a single string, to reduce the number of relocations when a shared
- as this is assumed for handling case independence. */
+ 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
- static const char *const posix_names[] = {
+ for handling case independence. */
-   "alpha", "lower", "upper",
-   "alnum", "ascii", "blank", "cntrl", "digit", "graph",
+ static const char posix_names[] =
-   "print", "punct", "space", "word",  "xdigit" };
+   "alpha\0"  "lower\0"  "upper\0"  "alnum\0"  "ascii\0"  "blank\0"
+   "cntrl\0"  "digit\0"  "graph\0"  "print\0"  "punct\0"  "space\0"
+   "word\0"   "xdigit";
  static const uschar posix_name_lengths[] = {
 , 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
-Line 207 
 static const int posix_class_maps[] = {
+Line 219 
 static const int posix_class_maps[] = {
  /* The texts of compile-time error messages. These are "char *" because they
  are passed to the outside world. Do not ever re-use any error number, because
  they are documented. Always add a new error instead. Messages marked DEAD below
- are no longer used. */
+ are no longer used. This used to be a table of strings, but in order to reduce
+ the number of relocations needed when a shared library is loaded dynamically,
- static const char *error_texts[] = {
+ it is now one long string. We cannot use a table of offsets, because the
-   "no error",
+ lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
-   "\\ at end of pattern",
+ simply count through to the one we want - this isn't a performance issue
-   "\\c at end of pattern",
+ because these strings are used only when there is a compilation error. */
-   "unrecognized character follows \\",
-   "numbers out of order in {} quantifier",
+ static const char error_texts[] =
+   "no error\0"
+   "\\ at end of pattern\0"
+   "\\c at end of pattern\0"
+   "unrecognized character follows \\\0"
+   "numbers out of order in {} quantifier\0"
    /* 5 */
-   "number too big in {} quantifier",
+   "number too big in {} quantifier\0"
-   "missing terminating ] for character class",
+   "missing terminating ] for character class\0"
-   "invalid escape sequence in character class",
+   "invalid escape sequence in character class\0"
-   "range out of order in character class",
+   "range out of order in character class\0"
-   "nothing to repeat",
+   "nothing to repeat\0"
    /* 10 */
-   "operand of unlimited repeat could match the empty string",  /** DEAD **/
+   "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
-   "internal error: unexpected repeat",
+   "internal error: unexpected repeat\0"
-   "unrecognized character after (?",
+   "unrecognized character after (? or (?-\0"
-   "POSIX named classes are supported only within a class",
+   "POSIX named classes are supported only within a class\0"
-   "missing )",
+   "missing )\0"
    /* 15 */
-   "reference to non-existent subpattern",
+   "reference to non-existent subpattern\0"
-   "erroffset passed as NULL",
+   "erroffset passed as NULL\0"
-   "unknown option bit(s) set",
+   "unknown option bit(s) set\0"
-   "missing ) after comment",
+   "missing ) after comment\0"
-   "parentheses nested too deeply",  /** DEAD **/
+   "parentheses nested too deeply\0"  /** DEAD **/
    /* 20 */
-   "regular expression is too large",
+   "regular expression is too large\0"
-   "failed to get memory",
+   "failed to get memory\0"
-   "unmatched parentheses",
+   "unmatched parentheses\0"
-   "internal error: code overflow",
+   "internal error: code overflow\0"
-   "unrecognized character after (?<",
+   "unrecognized character after (?<\0"
    /* 25 */
-   "lookbehind assertion is not fixed length",
+   "lookbehind assertion is not fixed length\0"
-   "malformed number or name after (?(",
+   "malformed number or name after (?(\0"
-   "conditional group contains more than two branches",
+   "conditional group contains more than two branches\0"
-   "assertion expected after (?(",
+   "assertion expected after (?(\0"
-   "(?R or (?[+-]digits must be followed by )",
+   "(?R or (?[+-]digits must be followed by )\0"
    /* 30 */
-   "unknown POSIX class name",
+   "unknown POSIX class name\0"
-   "POSIX collating elements are not supported",
+   "POSIX collating elements are not supported\0"
-   "this version of PCRE is not compiled with PCRE_UTF8 support",
+   "this version of PCRE is not compiled with PCRE_UTF8 support\0"
-   "spare error",  /** DEAD **/
+   "spare error\0"  /** DEAD **/
-   "character value in \\x{...} sequence is too large",
+   "character value in \\x{...} sequence is too large\0"
    /* 35 */
-   "invalid condition (?(0)",
+   "invalid condition (?(0)\0"
-   "\\C not allowed in lookbehind assertion",
+   "\\C not allowed in lookbehind assertion\0"
-   "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
+   "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
-   "number after (?C is > 255",
+   "number after (?C is > 255\0"
-   "closing ) for (?C expected",
+   "closing ) for (?C expected\0"
    /* 40 */
-   "recursive call could loop indefinitely",
+   "recursive call could loop indefinitely\0"
-   "unrecognized character after (?P",
+   "unrecognized character after (?P\0"
-   "syntax error in subpattern name (missing terminator)",
+   "syntax error in subpattern name (missing terminator)\0"
-   "two named subpatterns have the same name",
+   "two named subpatterns have the same name\0"
-   "invalid UTF-8 string",
+   "invalid UTF-8 string\0"
    /* 45 */
-   "support for \\P, \\p, and \\X has not been compiled",
+   "support for \\P, \\p, and \\X has not been compiled\0"
-   "malformed \\P or \\p sequence",
+   "malformed \\P or \\p sequence\0"
-   "unknown property name after \\P or \\p",
+   "unknown property name after \\P or \\p\0"
-   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)",
+   "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
-   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")",
+   "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
    /* 50 */
-   "repeated subpattern is too long",    /** DEAD **/
+   "repeated subpattern is too long\0"    /** DEAD **/
-   "octal value is greater than \\377 (not in UTF-8 mode)",
+   "octal value is greater than \\377 (not in UTF-8 mode)\0"
-   "internal error: overran compiling workspace",
+   "internal error: overran compiling workspace\0"
-   "internal error: previously-checked referenced subpattern not found",
+   "internal error: previously-checked referenced subpattern not found\0"
-   "DEFINE group contains more than one branch",
+   "DEFINE group contains more than one branch\0"
    /* 55 */
-   "repeating a DEFINE group is not allowed",
+   "repeating a DEFINE group is not allowed\0"
-   "inconsistent NEWLINE options",
+   "inconsistent NEWLINE options\0"
-   "\\g is not followed by a braced name or an optionally braced non-zero number",
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
-   "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number",
+   "a numbered reference must not be zero\0"
-   "(*VERB) with an argument is not supported",
+   "(*VERB) with an argument is not supported\0"
    /* 60 */
-   "(*VERB) not recognized"
+   "(*VERB) not recognized\0"
- };
+   "number is too big\0"
+   "subpattern name expected\0"
+   "digit expected after (?+\0"
+   "] is an invalid data character in JavaScript compatibility mode";
  /* Table to identify digits and hex digits. This is used when compiling
-Line 419 
 static BOOL
+Line 439 
 static BOOL
  /*************************************************
+ *            Find an error text                  *
+ *************************************************/
+ /* The error texts are now all in one long string, to save on relocations. As
+ some of the text is of unknown length, we can't use a table of offsets.
+ Instead, just count through the strings. This is not a performance issue
+ because it happens only when there has been a compilation error.
+ Argument:   the error number
+ Returns:    pointer to the error string
+ */
+ static const char *
+ find_error_text(int n)
+ {
+ const char *s = error_texts;
+ for (; n > 0; n--) while (*s++ != 0);
+ return s;
+ }
+ /*************************************************
  *            Handle escapes                      *
  *************************************************/
-Line 439 
 Arguments:
+Line 481 
 Arguments:
  Returns:         zero or positive => a data character
                   negative => a special escape sequence
-                  on error, errorptr is set
+                  on error, errorcodeptr is set
  */
  static int
-Line 457 
 ptr--;                            /* Set
+Line 499 
 ptr--;                            /* Set
  if (c == 0) *errorcodeptr = ERR1;
- /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
+ /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
- a table. A non-zero result is something that can be returned immediately.
+ in a table. A non-zero result is something that can be returned immediately.
  Otherwise further processing may be required. */
  #ifndef EBCDIC  /* ASCII coding */
- else if (c < '0' || c > 'z') {}                           /* Not alphameric */
+ else if (c < '0' || c > 'z') {}                           /* Not alphanumeric */
  else if ((i = escapes[c - '0']) != 0) c = i;
  #else           /* EBCDIC coding */
- else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */
+ else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
  else if ((i = escapes[c - 0x48]) != 0)  c = i;
  #endif
-Line 490 
 else
+Line 532 
 else
      *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by a number, either plain or braced. If positive, it
+     /* \g must be followed by one of a number of specific things:
-     is an absolute backreference. If negative, it is a relative backreference.
-     This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a
+     (1) A number, either plain or braced. If positive, it is an absolute
-     reference to a named group. This is part of Perl's movement towards a
+     backreference. If negative, it is a relative backreference. This is a Perl
-     unified syntax for back references. As this is synonymous with \k{name}, we
+.10 feature.
-     fudge it up by pretending it really was \k. */
+     (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
+     is part of Perl's movement towards a unified syntax for back references. As
+     this is synonymous with \k{name}, we fudge it up by pretending it really
+     was \k.
+     (3) For Oniguruma compatibility we also support \g followed by a name or a
+     number either in angle brackets or in single quotes. However, these are
+     (possibly recursive) subroutine calls, _not_ backreferences. Just return
+     the -ESC_g code (cf \k). */
      case 'g':
+     if (ptr[1] == '<' || ptr[1] == '\'')
+       {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
      if (ptr[1] == '{')
        {
        const uschar *p;
-Line 524 
 else
+Line 583 
 else
      while ((digitab[ptr[1]] & ctype_digit) != 0)
        c = c * 10 + *(++ptr) - '0';
-     if (c == 0 || (braced && *(++ptr) != '}'))
+     if (c < 0)   /* Integer overflow */
+       {
+       *errorcodeptr = ERR61;
+       break;
+       }
+     if (braced && *(++ptr) != '}')
        {
        *errorcodeptr = ERR57;
-       return 0;
+       break;
+       }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
        }
      if (negated)
-Line 535 
 else
+Line 606 
 else
        if (c > bracount)
          {
          *errorcodeptr = ERR15;
-         return 0;
+         break;
          }
        c = bracount - (c - 1);
        }
-Line 564 
 else
+Line 635 
 else
        c -= '0';
        while ((digitab[ptr[1]] & ctype_digit) != 0)
          c = c * 10 + *(++ptr) - '0';
+       if (c < 0)    /* Integer overflow */
+         {
+         *errorcodeptr = ERR61;
+         break;
+         }
        if (c < 10 || c <= bracount)
          {
          c = -(ESC_REF + c);
-Line 659 
 else
+Line 735 
 else
      if (c == 0)
        {
        *errorcodeptr = ERR2;
-       return 0;
+       break;
        }
  #ifndef EBCDIC  /* ASCII coding */
-Line 672 
 else
+Line 748 
 else
      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 764 
 top = _pcre_utt_size;
+Line 840 
 top = _pcre_utt_size;
  while (bot < top)
    {
    i = (bot + top) >> 1;
-   c = strcmp(name, _pcre_utt[i].name);
+   c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
    if (c == 0)
      {
      *dptr = _pcre_utt[i].value;
-Line 900 
 be terminated by '>' because that is che
+Line 976 
 be terminated by '>' because that is che
  Arguments:
    ptr          current position in the pattern
-   count        current count of capturing parens so far encountered
+   cd           compile background data
    name         name to seek, or NULL if seeking a numbered subpattern
    lorn         name length, or subpattern number if name is NULL
    xmode        TRUE if we are in /x mode
-Line 909 
 Returns:       the number of the named s
+Line 985 
 Returns:       the number of the named s
  */
  static int
- find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
+ find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,
    BOOL xmode)
  {
  const uschar *thisname;
+ int count = cd->bracount;
  for (; *ptr != 0; ptr++)
    {
-Line 932 
 for (; *ptr != 0; ptr++)
+Line 1009 
 for (; *ptr != 0; ptr++)
      continue;
      }
-   /* Skip over character classes */
+   /* Skip over character classes; this logic must be similar to the way they
+   are handled for real. If the first character is '^', skip it. Also, if the
+   first few characters (either before or after ^) are \Q\E or \E we skip them
+   too. This makes for compatibility with Perl. */
    if (*ptr == '[')
      {
+     BOOL negate_class = FALSE;
+     for (;;)
+       {
+       int c = *(++ptr);
+       if (c == '\\')
+         {
+         if (ptr[1] == 'E') ptr++;
+           else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
+             else break;
+         }
+       else if (!negate_class && c == '^')
+         negate_class = TRUE;
+       else break;
+       }
+     /* If the next character is ']', it is a data character that must be
+     skipped, except in JavaScript compatibility mode. */
+     if (ptr[1] == ']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
+       ptr++;
      while (*(++ptr) != ']')
        {
+       if (*ptr == 0) return -1;
        if (*ptr == '\\')
          {
          if (*(++ptr) == 0) return -1;
-Line 1093 
 for (;;)
+Line 1195 
 for (;;)
    {
    int d;
    register int op = *cc;
    switch (op)
      {
      case OP_CBRA:
-Line 1182 
 for (;;)
+Line 1283 
 for (;;)
      case OP_TYPEEXACT:
      branchlength += GET2(cc,1);
+     if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
      cc += 4;
      break;
-Line 1199 
 for (;;)
+Line 1301 
 for (;;)
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      branchlength++;
      cc++;
      break;
-Line 1290 
 for (;;)
+Line 1393 
 for (;;)
      code += _pcre_OP_lengths[c];
      }
-   /* In UTF-8 mode, opcodes that are followed by a character may be followed by
+   /* Otherwise, we can get the item's length from the table, except that for
-   a multi-byte character. The length in the table is a minimum, so we have to
+   repeated character types, we have to test for \p and \P, which have an extra
-   arrange to skip the extra bytes. */
+   two bytes of parameters. */
    else
      {
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       case OP_TYPEPOSUPTO:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
+       }
+     /* Add in the fixed length from the table */
      code += _pcre_OP_lengths[c];
+   /* In UTF-8 mode, opcodes that are followed by a character may be followed by
+   a multi-byte character. The length in the table is a minimum, so we have to
+   arrange to skip the extra bytes. */
  #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
-Line 1354 
 for (;;)
+Line 1486 
 for (;;)
    if (c == OP_XCLASS) code += GET(code, 1);
-   /* Otherwise, we get the item's length from the table. In UTF-8 mode, opcodes
+   /* Otherwise, we can get the item's length from the table, except that for
-   that are followed by a character may be followed by a multi-byte character.
+   repeated character types, we have to test for \p and \P, which have an extra
-   The length in the table is a minimum, so we have to arrange to skip the extra
+   two bytes of parameters. */
-   bytes. */
    else
      {
+     switch(c)
+       {
+       case OP_TYPESTAR:
+       case OP_TYPEMINSTAR:
+       case OP_TYPEPLUS:
+       case OP_TYPEMINPLUS:
+       case OP_TYPEQUERY:
+       case OP_TYPEMINQUERY:
+       case OP_TYPEPOSSTAR:
+       case OP_TYPEPOSPLUS:
+       case OP_TYPEPOSQUERY:
+       if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+       break;
+       case OP_TYPEPOSUPTO:
+       case OP_TYPEUPTO:
+       case OP_TYPEMINUPTO:
+       case OP_TYPEEXACT:
+       if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+       break;
+       }
+     /* Add in the fixed length from the table */
      code += _pcre_OP_lengths[c];
+     /* In UTF-8 mode, opcodes that are followed by a character may be followed
+     by a multi-byte character. The length in the table is a minimum, so we have
+     to arrange to skip the extra bytes. */
  #ifdef SUPPORT_UTF8
      if (utf8) switch(c)
        {
-Line 1398 
 for (;;)
+Line 1558 
 for (;;)
  can match the empty string or not. It is called from could_be_empty()
  below and from compile_branch() when checking for an unlimited repeat of a
  group that can match nothing. Note that first_significant_code() skips over
- assertions. If we hit an unclosed bracket, we return "empty" - this means we've
+ backward and negative forward assertions when its final argument is TRUE. If we
- struck an inner bracket whose current branch will already have been scanned.
+ hit an unclosed bracket, we return "empty" - this means we've struck an inner
+ bracket whose current branch will already have been scanned.
  Arguments:
    code        points to start of search
-Line 1421 
 for (code = first_significant_code(code
+Line 1582 
 for (code = first_significant_code(code
    c = *code;
+   /* Skip over forward assertions; the other assertions are skipped by
+   first_significant_code() with a TRUE final argument. */
+   if (c == OP_ASSERT)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
    /* Groups with zero repeats can of course be empty; skip them. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO)
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
      {
      code += _pcre_OP_lengths[c];
      do code += GET(code, 1); while (*code == OP_ALT);
-Line 1457 
 for (code = first_significant_code(code
+Line 1628 
 for (code = first_significant_code(code
    switch (c)
      {
-     /* Check for quantifiers after a class */
+     /* Check for quantifiers after a class. XCLASS is used for classes that
+     cannot be represented just by a bit map. This includes negated single
+     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
+     actual length is stored in the compiled code, so we must update "code"
+     here. */
  #ifdef SUPPORT_UTF8
      case OP_XCLASS:
-     ccode = code + GET(code, 1);
+     ccode = code += GET(code, 1);
      goto CHECK_CLASS_REPEAT;
  #endif
-Line 1505 
 for (code = first_significant_code(code
+Line 1680 
 for (code = first_significant_code(code
      case OP_NOT_WORDCHAR:
      case OP_WORDCHAR:
      case OP_ANY:
+     case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
      case OP_CHARNC:
-Line 1523 
 for (code = first_significant_code(code
+Line 1699 
 for (code = first_significant_code(code
      case OP_TYPEEXACT:
      return FALSE;
+     /* These are going to continue, as they may be empty, but we have to
+     fudge the length for the \p and \P cases. */
+     case OP_TYPESTAR:
+     case OP_TYPEMINSTAR:
+     case OP_TYPEPOSSTAR:
+     case OP_TYPEQUERY:
+     case OP_TYPEMINQUERY:
+     case OP_TYPEPOSQUERY:
+     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
+     break;
+     /* Same for these */
+     case OP_TYPEUPTO:
+     case OP_TYPEMINUPTO:
+     case OP_TYPEPOSUPTO:
+     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
+     break;
      /* End of branch */
      case OP_KET:
-Line 1592 
 return TRUE;
+Line 1788 
 return TRUE;
  *************************************************/
  /* This function is called when the sequence "[:" or "[." or "[=" is
- encountered in a character class. It checks whether this is followed by an
+ encountered in a character class. It checks whether this is followed by a
- optional ^ and then a sequence of letters, terminated by a matching ":]" or
+ sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
- ".]" or "=]".
+ reach an unescaped ']' without the special preceding character, return FALSE.
+ Originally, this function only recognized a sequence of letters between the
+ terminators, but it seems that Perl recognizes any sequence of characters,
+ though of course unknown POSIX names are subsequently rejected. Perl gives an
+ "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
+ didn't consider this to be a POSIX class. Likewise for [:1234:].
+ The problem in trying to be exactly like Perl is in the handling of escapes. We
+ have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
+ class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
+ below handles the special case of \], but does not try to do any other escape
+ processing. This makes it different from Perl for cases such as [:l\ower:]
+ where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
+ "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
+ I think.
- Argument:
+ Arguments:
    ptr      pointer to the initial [
    endptr   where to return the end pointer
-   cd       pointer to compile data
  Returns:   TRUE or FALSE
  */
  static BOOL
- check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)
+ check_posix_syntax(const uschar *ptr, const uschar **endptr)
  {
  int terminator;          /* Don't combine these lines; the Solaris cc */
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
- if (*(++ptr) == '^') ptr++;
+ for (++ptr; *ptr != 0; ptr++)
- while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
- if (*ptr == terminator && ptr[1] == ']')
    {
-   *endptr = ptr;
+   if (*ptr == '\\' && ptr[1] == ']') ptr++; else
-   return TRUE;
+     {
+     if (*ptr == ']') return FALSE;
+     if (*ptr == terminator && ptr[1] == ']')
+       {
+       *endptr = ptr;
+       return TRUE;
+       }
+     }
    }
  return FALSE;
  }
-Line 1639 
 Returns:     a value representing the na
+Line 1854 
 Returns:     a value representing the na
  static int
  check_posix_name(const uschar *ptr, int len)
  {
+ const char *pn = posix_names;
  register int yield = 0;
  while (posix_name_lengths[yield] != 0)
    {
    if (len == posix_name_lengths[yield] &&
-     strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;
+     strncmp((const char *)ptr, pn, len) == 0) return yield;
+   pn += posix_name_lengths[yield] + 1;
    yield++;
    }
  return -1;
-Line 1658 
 return -1;
+Line 1875 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That one of the jobs of this function. Before it is called,
+ have their offsets adjusted. That one of the jobs of this function. Before it
- the partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
  This function has been extended with the possibility of forward references for
  recursions and subroutine calls. It must also check the list of such references
-Line 1685 
 adjust_recurse(uschar *group, int adjust
+Line 1903 
 adjust_recurse(uschar *group, int adjust
    uschar *save_hwm)
  {
  uschar *ptr = group;
  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
    {
    int offset;
-Line 1946 
 if (next >= 0) switch(op_code)
+Line 2165 
 if (next >= 0) switch(op_code)
    /* For OP_NOT, "item" must be a single-byte character. */
    case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
    if (item == next) return TRUE;
    if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
-Line 2209 
 uschar classbits[32];
+Line 2427 
 uschar classbits[32];
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
-Line 2248 
 req_caseopt = ((options & PCRE_CASELESS)
+Line 2467 
 req_caseopt = ((options & PCRE_CASELESS)
  for (;; ptr++)
    {
    BOOL negate_class;
+   BOOL should_flip_negation;
    BOOL possessive_quantifier;
    BOOL is_quantifier;
    BOOL is_recurse;
-Line 2447 
 for (;; ptr++)
+Line 2667 
 for (;; ptr++)
      zerofirstbyte = firstbyte;
      zeroreqbyte = reqbyte;
      previous = code;
-     *code++ = OP_ANY;
+     *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
      break;
-Line 2462 
 for (;; ptr++)
+Line 2682 
 for (;; ptr++)
      opcode is compiled. It may optionally have a bit map for characters < 256,
      but those above are are explicitly listed afterwards. A flag byte tells
      whether the bitmap is present, and whether this is a negated class or not.
-     */
+     In JavaScript compatibility mode, an isolated ']' causes an error. In
+     default (Perl) mode, it is treated as a data character. */
+     case ']':
+     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *errorcodeptr = ERR64;
+       goto FAILED;
+       }
+     goto NORMAL_CHAR;
      case '[':
      previous = code;
-Line 2471 
 for (;; ptr++)
+Line 2701 
 for (;; ptr++)
      they are encountered at the top level, so we'll do that too. */
      if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
-         check_posix_syntax(ptr, &tempptr, cd))
+         check_posix_syntax(ptr, &tempptr))
        {
        *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
        goto FAILED;
-Line 2496 
 for (;; ptr++)
+Line 2726 
 for (;; ptr++)
        else break;
        }
+     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
+     an initial ']' is taken as a data character -- the code below handles
+     that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
+     [^] must match any character, so generate OP_ALLANY. */
+     if (c ==']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *code++ = negate_class? OP_ALLANY : OP_FAIL;
+       if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       zerofirstbyte = firstbyte;
+       break;
+       }
+     /* If a class contains a negative special such as \S, we need to flip the
+     negation flag at the end, so that support for characters > 255 works
+     correctly (they are all included in the class). */
+     should_flip_negation = FALSE;
      /* Keep a count of chars with values < 256 so that we can optimize the case
      of just a single character (as long as it's < 256). However, For higher
      valued UTF-8 characters, we don't yet do any optimization. */
-Line 2513 
 for (;; ptr++)
+Line 2762 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
      class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 2528 
 for (;; ptr++)
+Line 2778 
 for (;; ptr++)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+       data and reset the pointer. This is so that very large classes that
+       contain a zillion UTF-8 characters no longer overwrite the work space
+       (which is on the stack). */
+       if (lengthptr != NULL)
+         {
+         *lengthptr += class_utf8data - class_utf8data_base;
+         class_utf8data = class_utf8data_base;
+         }
  #endif
        /* Inside \Q...\E everything is literal except \E */
-Line 2551 
 for (;; ptr++)
+Line 2813 
 for (;; ptr++)
        if (c == '[' &&
            (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
-           check_posix_syntax(ptr, &tempptr, cd))
+           check_posix_syntax(ptr, &tempptr))
          {
          BOOL local_negate = FALSE;
          int posix_class, taboffset, tabopt;
-Line 2568 
 for (;; ptr++)
+Line 2830 
 for (;; ptr++)
          if (*ptr == '^')
            {
            local_negate = TRUE;
+           should_flip_negation = TRUE;  /* Note negative special */
            ptr++;
            }
-Line 2642 
 for (;; ptr++)
+Line 2905 
 for (;; ptr++)
          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
          if (*errorcodeptr != 0) goto FAILED;
-         if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */
+         if (-c == ESC_b) c = '\b';       /* \b is backspace in a class */
          else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */
          else if (-c == ESC_R) c = 'R';   /* \R is literal R in a class */
          else if (-c == ESC_Q)            /* Handle start of quoted string */
-Line 2654 
 for (;; ptr++)
+Line 2917 
 for (;; ptr++)
            else inescq = TRUE;
            continue;
            }
+         else if (-c == ESC_E) continue;  /* Ignore orphan \E */
          if (c < 0)
            {
-Line 2669 
 for (;; ptr++)
+Line 2933 
 for (;; ptr++)
              continue;
              case ESC_D:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
              continue;
-Line 2677 
 for (;; ptr++)
+Line 2942 
 for (;; ptr++)
              continue;
              case ESC_W:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
-Line 2686 
 for (;; ptr++)
+Line 2952 
 for (;; ptr++)
              continue;
              case ESC_S:
+             should_flip_negation = TRUE;
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
              continue;
-             case ESC_E: /* Perl ignores an orphan \E */
-             continue;
              default:    /* Not recognized; fall through */
              break;      /* Need "default" setting to stop compiler warning. */
              }
-Line 2880 
 for (;; ptr++)
+Line 3144 
 for (;; ptr++)
        oldptr = ptr;
+       /* Remember \r or \n */
+       if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF;
+       /* Check for range */
        if (!inescq && ptr[1] == '-')
          {
          int d;
-Line 2921 
 for (;; ptr++)
+Line 3191 
 for (;; ptr++)
            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
            if (*errorcodeptr != 0) goto FAILED;
-           /* \b is backslash; \X is literal X; \R is literal R; any other
+           /* \b is backspace; \X is literal X; \R is literal R; any other
            special means the '-' was literal */
            if (d < 0)
-Line 2947 
 for (;; ptr++)
+Line 3217 
 for (;; ptr++)
          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */
+         /* Remember \r or \n */
+         if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF;
          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
          matching, we have to use an XCLASS with extra data items. Caseless
          matching for characters > 127 is available only if UCP support is
-Line 3100 
 for (;; ptr++)
+Line 3374 
 for (;; ptr++)
        goto FAILED;
        }
+ /* This code has been disabled because it would mean that \s counts as
+ an explicit \r or \n reference, and that's not really what is wanted. Now
+ we set the flag only if there is a literal "\r" or "\n" in the class. */
+ #if 0
+     /* Remember whether \r or \n are in this class */
+     if (negate_class)
+       {
+       if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
+       }
+     else
+       {
+       if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
+       }
+ #endif
      /* If class_charcount is 1, we saw precisely one character whose value is
-     less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we
+     less than 256. As long as there were no characters >= 128 and there was no
-     can optimize the negative case only if there were no characters >= 128
+     use of \p or \P, in other words, no use of any XCLASS features, we can
-     because OP_NOT and the related opcodes like OP_NOTSTAR operate on
+     optimize.
-     single-bytes only. This is an historical hangover. Maybe one day we can
-     tidy these opcodes to handle multi-byte characters.
+     In UTF-8 mode, we can optimize the negative case only if there were no
+     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
+     operate on single-bytes only. This is an historical hangover. Maybe one day
+     we can tidy these opcodes to handle multi-byte characters.
      The optimization throws away the bit map. We turn the item into a
 -character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
-Line 3115 
 for (;; ptr++)
+Line 3411 
 for (;; ptr++)
      reqbyte, save the previous value for reinstating. */
  #ifdef SUPPORT_UTF8
-     if (class_charcount == 1 &&
+     if (class_charcount == 1 && !class_utf8 &&
-           (!utf8 ||
+       (!utf8 || !negate_class || class_lastchar < 128))
-           (!class_utf8 && (!negate_class || class_lastchar < 128))))
  #else
      if (class_charcount == 1)
  #endif
-Line 3161 
 for (;; ptr++)
+Line 3455 
 for (;; ptr++)
      zeroreqbyte = reqbyte;
      /* If there are characters with values > 255, we have to compile an
-     extended class, with its own opcode. If there are no characters < 256,
+     extended class, with its own opcode, unless there was a negated special
-     we can omit the bitmap in the actual compiled code. */
+     such as \S in the class, because in that case all characters > 255 are in
+     the class, so any that were explicitly given as well can be ignored. If
+     (when there are explicit characters > 255 that must be listed) there are no
+     characters < 256, we can omit the bitmap in the actual compiled code. */
  #ifdef SUPPORT_UTF8
-     if (class_utf8)
+     if (class_utf8 && !should_flip_negation)
        {
        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
        *code++ = OP_XCLASS;
-Line 3191 
 for (;; ptr++)
+Line 3488 
 for (;; ptr++)
        }
  #endif
-     /* If there are no characters > 255, negate the 32-byte map if necessary,
+     /* If there are no characters > 255, set the opcode to OP_CLASS or
-     and copy it into the code vector. If this is the first thing in the branch,
+     OP_NCLASS, depending on whether the whole class was negated and whether
-     there can be no first char setting, whatever the repeat count. Any reqbyte
+     there were negative specials such as \S in the class. Then copy the 32-byte
-     setting must remain unchanged after any kind of repeat. */
+     map into the code vector, negating it if necessary. */
+     *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
      if (negate_class)
        {
-       *code++ = OP_NCLASS;
        if (lengthptr == NULL)    /* Save time in the pre-compile phase */
          for (c = 0; c < 32; c++) code[c] = ~classbits[c];
        }
      else
        {
-       *code++ = OP_CLASS;
        memcpy(code, classbits, 32);
        }
      code += 32;
-Line 3390 
 for (;; ptr++)
+Line 3686 
 for (;; ptr++)
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->nopartial = TRUE;
+       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
        /* Combine the op_type with the repeat_type */
-Line 3540 
 for (;; ptr++)
+Line 3836 
 for (;; ptr++)
        /* All real repeats make it impossible to handle partial matching (maybe
        one day we will be able to remove this restriction). */
-       if (repeat_max != 1) cd->nopartial = TRUE;
+       if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
        if (repeat_min == 0 && repeat_max == -1)
          *code++ = OP_CRSTAR + repeat_type;
-Line 3598 
 for (;; ptr++)
+Line 3894 
 for (;; ptr++)
        if (repeat_min == 0)
          {
-         /* If the maximum is also zero, we just omit the group from the output
+         /* If the maximum is also zero, we used to just omit the group from the
-         altogether. */
+         output altogether, like this:
-         if (repeat_max == 0)
+         ** if (repeat_max == 0)
-           {
+         **   {
-           code = previous;
+         **   code = previous;
-           goto END_REPEAT;
+         **   goto END_REPEAT;
-           }
+         **   }
-         /* If the maximum is 1 or unlimited, we just have to stick in the
+         However, that fails when a group is referenced as a subroutine from
-         BRAZERO and do no more at this point. However, we do need to adjust
+         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
-         any OP_RECURSE calls inside the group that refer to the group itself or
+         so that it is skipped on execution. As we don't have a list of which
-         any internal or forward referenced group, because the offset is from
+         groups are referenced, we cannot do this selectively.
-         the start of the whole regex. Temporarily terminate the pattern while
-         doing this. */
+         If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
+         and do no more at this point. However, we do need to adjust any
+         OP_RECURSE calls inside the group that refer to the group itself or any
+         internal or forward referenced group, because the offset is from the
+         start of the whole regex. Temporarily terminate the pattern while doing
+         this. */
-         if (repeat_max <= 1)
+         if (repeat_max <= 1)    /* Covers 0, 1, and unlimited */
            {
            *code = OP_END;
            adjust_recurse(previous, 1, utf8, cd, save_hwm);
            memmove(previous+1, previous, len);
            code++;
+           if (repeat_max == 0)
+             {
+             *previous++ = OP_SKIPZERO;
+             goto END_REPEAT;
+             }
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 3814 
 for (;; ptr++)
+Line 4120 
 for (;; ptr++)
          }
        }
+     /* If previous is OP_FAIL, it was generated by an empty class [] in
+     JavaScript mode. The other ways in which OP_FAIL can be generated, that is
+     by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
+     error above. We can just ignore the repeat in JS case. */
+     else if (*previous == OP_FAIL) goto END_REPEAT;
      /* Else there's some kind of shambles */
      else
-Line 3840 
 for (;; ptr++)
+Line 4153 
 for (;; ptr++)
        int len;
        if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
            *tempcode == OP_NOTEXACT)
-         tempcode += _pcre_OP_lengths[*tempcode];
+         tempcode += _pcre_OP_lengths[*tempcode] +
+           ((*tempcode == OP_TYPEEXACT &&
+              (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
        len = code - tempcode;
        if (len > 0) switch (*tempcode)
          {
-Line 3892 
 for (;; ptr++)
+Line 4207 
 for (;; ptr++)
      bravalue = OP_CBRA;
      save_hwm = cd->hwm;
      reset_bracount = FALSE;
      /* First deal with various "verbs" that can be introduced by '*'. */
      if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
        {
        int i, namelen;
+       const char *vn = verbnames;
        const uschar *name = ++ptr;
        previous = NULL;
        while ((cd->ctypes[*++ptr] & ctype_letter) != 0);
        if (*ptr == ':')
          {
          *errorcodeptr = ERR59;   /* Not supported */
          goto FAILED;
          }
        if (*ptr != ')')
          {
          *errorcodeptr = ERR60;
          goto FAILED;
          }
        namelen = ptr - name;
        for (i = 0; i < verbcount; i++)
          {
          if (namelen == verbs[i].len &&
-             strncmp((char *)name, verbs[i].name, namelen) == 0)
+             strncmp((char *)name, vn, namelen) == 0)
            {
            *code = verbs[i].op;
            if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
            break;
            }
-         }
+         vn += verbs[i].len + 1;
-       if (i < verbcount) continue;
+         }
+       if (i < verbcount) continue;
        *errorcodeptr = ERR60;
        goto FAILED;
        }
      /* Deal with the extended parentheses; all are introduced by '?', and the
      appearance of any of them means that this is not a capturing group. */
-Line 4065 
 for (;; ptr++)
+Line 4382 
 for (;; ptr++)
              *errorcodeptr = ERR58;
              goto FAILED;
              }
-           if (refsign == '-')
+           recno = (refsign == '-')?
+             cd->bracount - recno + 1 : recno +cd->bracount;
+           if (recno <= 0 || recno > cd->final_bracount)
              {
-             recno = cd->bracount - recno + 1;
+             *errorcodeptr = ERR15;
-             if (recno <= 0)
+             goto FAILED;
-               {
-               *errorcodeptr = ERR15;
-               goto FAILED;
-               }
              }
-           else recno += cd->bracount;
            PUT2(code, 2+LINK_SIZE, recno);
            break;
            }
-Line 4099 
 for (;; ptr++)
+Line 4413 
 for (;; ptr++)
          /* Search the pattern for a forward reference */
-         else if ((i = find_parens(ptr, cd->bracount, name, namelen,
+         else if ((i = find_parens(ptr, cd, name, namelen,
                          (options & PCRE_EXTENDED) != 0)) > 0)
            {
            PUT2(code, 2+LINK_SIZE, i);
-Line 4146 
 for (;; ptr++)
+Line 4460 
 for (;; ptr++)
            skipbytes = 1;
            }
-         /* Check for the "name" actually being a subpattern number. */
+         /* Check for the "name" actually being a subpattern number. We are
+         in the second pass here, so final_bracount is set. */
-         else if (recno > 0)
+         else if (recno > 0 && recno <= cd->final_bracount)
            {
            PUT2(code, 2+LINK_SIZE, recno);
            }
-Line 4176 
 for (;; ptr++)
+Line 4491 
 for (;; ptr++)
          if (*ptr == ')')          /* Optimize (?!) */
            {
            *code++ = OP_FAIL;
            previous = NULL;
            continue;
            }
          bravalue = OP_ASSERT_NOT;
          break;
-Line 4342 
 for (;; ptr++)
+Line 4657 
 for (;; ptr++)
          /* We come here from the Python syntax above that handles both
          references (?P=name) and recursion (?P>name), as well as falling
-         through from the Perl recursion syntax (?&name). */
+         through from the Perl recursion syntax (?&name). We also come here from
+         the Perl \k<name> or \k'name' back reference syntax and the \k{name}
+         .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
          NAMED_REF_OR_RECURSE:
          name = ++ptr;
-Line 4354 
 for (;; ptr++)
+Line 4671 
 for (;; ptr++)
          if (lengthptr != NULL)
            {
+           if (namelen == 0)
+             {
+             *errorcodeptr = ERR62;
+             goto FAILED;
+             }
            if (*ptr != terminator)
              {
              *errorcodeptr = ERR42;
-Line 4367 
 for (;; ptr++)
+Line 4689 
 for (;; ptr++)
            recno = 0;
            }
-         /* In the real compile, seek the name in the table */
+         /* In the real compile, seek the name in the table. We check the name
+         first, and then check that we have reached the end of the name in the
+         table. That way, if the name that is longer than any in the table,
+         the comparison will fail without reading beyond the table entry. */
          else
            {
            slot = cd->name_table;
            for (i = 0; i < cd->names_found; i++)
              {
-             if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
+             if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
+                 slot[2+namelen] == 0)
+               break;
              slot += cd->name_entry_size;
              }
-Line 4383 
 for (;; ptr++)
+Line 4710 
 for (;; ptr++)
              recno = GET2(slot, 0);
              }
            else if ((recno =                /* Forward back reference */
-                     find_parens(ptr, cd->bracount, name, namelen,
+                     find_parens(ptr, cd, name, namelen,
                        (options & PCRE_EXTENDED) != 0)) <= 0)
              {
              *errorcodeptr = ERR15;
-Line 4410 
 for (;; ptr++)
+Line 4737 
 for (;; ptr++)
          case '5': case '6': case '7': case '8': case '9':   /* subroutine */
            {
            const uschar *called;
+           terminator = ')';
+           /* Come here from the \g<...> and \g'...' code (Oniguruma
+           compatibility). However, the syntax has been checked to ensure that
+           the ... are a (signed) number, so that neither ERR63 nor ERR29 will
+           be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
+           ever be taken. */
+           HANDLE_NUMERICAL_RECURSION:
-           if ((refsign = *ptr) == '+') ptr++;
+           if ((refsign = *ptr) == '+')
+             {
+             ptr++;
+             if ((digitab[*ptr] & ctype_digit) == 0)
+               {
+               *errorcodeptr = ERR63;
+               goto FAILED;
+               }
+             }
            else if (refsign == '-')
              {
              if ((digitab[ptr[1]] & ctype_digit) == 0)
-Line 4423 
 for (;; ptr++)
+Line 4767 
 for (;; ptr++)
            while((digitab[*ptr] & ctype_digit) != 0)
              recno = recno * 10 + *ptr++ - '0';
-           if (*ptr != ')')
+           if (*ptr != terminator)
              {
              *errorcodeptr = ERR29;
              goto FAILED;
-Line 4476 
 for (;; ptr++)
+Line 4820 
 for (;; ptr++)
              if (called == NULL)
                {
-               if (find_parens(ptr, cd->bracount, NULL, recno,
+               if (find_parens(ptr, cd, NULL, recno,
-                    (options & PCRE_EXTENDED) != 0) < 0)
+                     (options & PCRE_EXTENDED) != 0) < 0)
                  {
                  *errorcodeptr = ERR15;
                  goto FAILED;
-Line 4537 
 for (;; ptr++)
+Line 4881 
 for (;; ptr++)
              case 'J':    /* Record that it changed in the external options */
              *optset |= PCRE_DUPNAMES;
-             cd->external_options |= PCRE_JCHANGED;
+             cd->external_flags |= PCRE_JCHANGED;
              break;
              case 'i': *optset |= PCRE_CASELESS; break;
-Line 4728 
 for (;; ptr++)
+Line 5072 
 for (;; ptr++)
        goto FAILED;
        }
-     /* In the pre-compile phase, update the length by the length of the nested
+     /* In the pre-compile phase, update the length by the length of the group,
-     group, less the brackets at either end. Then reduce the compiled code to
+     less the brackets at either end. Then reduce the compiled code to just a
-     just the brackets so that it doesn't use much memory if it is duplicated by
+     set of non-capturing brackets so that it doesn't use much memory if it is
-     a quantifier. */
+     duplicated by a quantifier.*/
      if (lengthptr != NULL)
        {
-Line 4741 
 for (;; ptr++)
+Line 5085 
 for (;; ptr++)
          goto FAILED;
          }
        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
-       code++;
+       *code++ = OP_BRA;
        PUTINC(code, 0, 1 + LINK_SIZE);
        *code++ = OP_KET;
        PUTINC(code, 0, 1 + LINK_SIZE);
+       break;    /* No need to waste time with special character handling */
        }
      /* Otherwise update the main code pointer to the end of the group. */
-     else code = tempcode;
+     code = tempcode;
      /* For a DEFINE group, required and first character settings are not
      relevant. */
-Line 4846 
 for (;; ptr++)
+Line 5191 
 for (;; ptr++)
        zerofirstbyte = firstbyte;
        zeroreqbyte = reqbyte;
+       /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
+       is a subroutine call by number (Oniguruma syntax). In fact, the value
+       -ESC_g is returned only for these cases. So we don't need to check for <
+       or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
+       -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
+       that is a synonym for a named back reference). */
+       if (-c == ESC_g)
+         {
+         const uschar *p;
+         save_hwm = cd->hwm;   /* Normally this is set when '(' is read */
+         terminator = (*(++ptr) == '<')? '>' : '\'';
+         /* These two statements stop the compiler for warning about possibly
+         unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
+         fact, because we actually check for a number below, the paths that
+         would actually be in error are never taken. */
+         skipbytes = 0;
+         reset_bracount = FALSE;
+         /* Test for a name */
+         if (ptr[1] != '+' && ptr[1] != '-')
+           {
+           BOOL isnumber = TRUE;
+           for (p = ptr + 1; *p != 0 && *p != terminator; p++)
+             {
+             if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
+             if ((cd->ctypes[*p] & ctype_word) == 0) break;
+             }
+           if (*p != terminator)
+             {
+             *errorcodeptr = ERR57;
+             break;
+             }
+           if (isnumber)
+             {
+             ptr++;
+             goto HANDLE_NUMERICAL_RECURSION;
+             }
+           is_recurse = TRUE;
+           goto NAMED_REF_OR_RECURSE;
+           }
+         /* Test a signed number in angle brackets or quotes. */
+         p = ptr + 2;
+         while ((digitab[*p] & ctype_digit) != 0) p++;
+         if (*p != terminator)
+           {
+           *errorcodeptr = ERR57;
+           break;
+           }
+         ptr++;
+         goto HANDLE_NUMERICAL_RECURSION;
+         }
        /* \k<name> or \k'name' is a back reference by name (Perl syntax).
        We also support \k{name} (.NET syntax) */
-Line 4953 
 for (;; ptr++)
+Line 5356 
 for (;; ptr++)
      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
+     /* Remember if \r or \n were seen */
+     if (mcbuffer[0] == '\r' || mcbuffer[0] == '\n')
+       cd->external_flags |= PCRE_HASCRORLF;
      /* Set the first and required bytes appropriately. If no previous first
      byte, set it from this character, but revert to none on a zero repeat.
      Otherwise, leave the firstbyte value alone, and don't change it on a zero
-Line 5347 
 do {
+Line 5755 
 do {
       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
       }
-    /* .* is not anchored unless DOTALL is set and it isn't in brackets that
+    /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
-    are or may be referenced. */
+    it isn't in brackets that are or may be referenced. */
     else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
-              op == OP_TYPEPOSSTAR) &&
+              op == OP_TYPEPOSSTAR))
-             (*options & PCRE_DOTALL) != 0)
       {
-      if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
+      if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
+        return FALSE;
       }
     /* Check for explicit anchoring */
-Line 5552 
 real_pcre *re;
+Line 5960 
 real_pcre *re;
  int length = 1;  /* For final END opcode */
  int firstbyte, reqbyte, newline;
  int errorcode = 0;
+ int skipatstart = 0;
  #ifdef SUPPORT_UTF8
  BOOL utf8;
  #endif
-Line 5570 
 to fill in forward references to subpatt
+Line 5979 
 to fill in forward references to subpatt
  uschar cworkspace[COMPILE_WORK_SIZE];
  /* Set this early so that early errors get offset 0. */
  ptr = (const uschar *)pattern;
-Line 5630 
 cd->fcc = tables + fcc_offset;
+Line 6038 
 cd->fcc = tables + fcc_offset;
  cd->cbits = tables + cbits_offset;
  cd->ctypes = tables + ctypes_offset;
+ /* Check for global one-time settings at the start of the pattern, and remember
+ the offset for later. */
+ while (ptr[skipatstart] == '(' && ptr[skipatstart+1] == '*')
+   {
+   int newnl = 0;
+   int newbsr = 0;
+   if (strncmp((char *)(ptr+skipatstart+2), "CR)", 3) == 0)
+     { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "LF)", 3)  == 0)
+     { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "CRLF)", 5)  == 0)
+     { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "ANY)", 4) == 0)
+     { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "ANYCRLF)", 8)  == 0)
+     { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "BSR_ANYCRLF)", 12) == 0)
+     { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
+   else if (strncmp((char *)(ptr+skipatstart+2), "BSR_UNICODE)", 12) == 0)
+     { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
+   if (newnl != 0)
+     options = (options & ~PCRE_NEWLINE_BITS) | newnl;
+   else if (newbsr != 0)
+     options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
+   else break;
+   }
+ /* Check validity of \R options. */
+ switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
+   {
+   case 0:
+   case PCRE_BSR_ANYCRLF:
+   case PCRE_BSR_UNICODE:
+   break;
+   default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
+   }
  /* Handle different types of newline. The three bits give seven cases. The
  current code allows for fixed one- or two-byte sequences, plus "any" and
  "anycrlf". */
- switch (options & (PCRE_NEWLINE_CRLF | PCRE_NEWLINE_ANY))
+ switch (options & PCRE_NEWLINE_BITS)
    {
-   case 0: newline = NEWLINE; break;   /* Compile-time default */
+   case 0: newline = NEWLINE; break;   /* Build-time default */
    case PCRE_NEWLINE_CR: newline = '\r'; break;
    case PCRE_NEWLINE_LF: newline = '\n'; break;
    case PCRE_NEWLINE_CR+
-Line 5689 
 to compile parts of the pattern into; th
+Line 6139 
 to compile parts of the pattern into; th
  no longer needed, so hopefully this workspace will never overflow, though there
  is a test for its doing so. */
- cd->bracount = 0;
+ cd->bracount = cd->final_bracount = 0;
  cd->names_found = 0;
  cd->name_entry_size = 0;
  cd->name_table = NULL;
-Line 5699 
 cd->hwm = cworkspace;
+Line 6149 
 cd->hwm = cworkspace;
  cd->start_pattern = (const uschar *)pattern;
  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
  cd->req_varyopt = 0;
- cd->nopartial = FALSE;
  cd->external_options = options;
+ cd->external_flags = 0;
  /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
  don't need to look at the result of the function here. The initial options have
-Line 5708 
 been put into the cd block so that they
+Line 6158 
 been put into the cd block so that they
  found within the regex right at the beginning. Bringing initial option settings
  outside can help speed up starting point checks. */
+ ptr += skipatstart;
  code = cworkspace;
  *code = OP_BRA;
  (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
-Line 5738 
 if (re == NULL)
+Line 6189 
 if (re == NULL)
    goto PCRE_EARLY_ERROR_RETURN;
    }
- /* Put in the magic number, and save the sizes, initial options, and character
+ /* Put in the magic number, and save the sizes, initial options, internal
- table pointer. NULL is used for the default character tables. The nullpad field
+ flags, and character table pointer. NULL is used for the default character
- is at the end; it's there to help in the case when a regex compiled on a system
+ tables. The nullpad field is at the end; it's there to help in the case when a
- with 4-byte pointers is run on another with 8-byte pointers. */
+ regex compiled on a system with 4-byte pointers is run on another with 8-byte
+ pointers. */
  re->magic_number = MAGIC_NUMBER;
  re->size = size;
  re->options = cd->external_options;
+ re->flags = cd->external_flags;
  re->dummy1 = 0;
  re->first_byte = 0;
  re->req_byte = 0;
-Line 5763 
 field. Reset the bracket count and the n
+Line 6216 
 field. Reset the bracket count and the n
  field; this time it's used for remembering forward references to subpatterns.
  */
+ cd->final_bracount = cd->bracount;  /* Save for checking forward references */
  cd->bracount = 0;
  cd->names_found = 0;
  cd->name_table = (uschar *)re + re->name_table_offset;
-Line 5770 
 codestart = cd->name_table + re->name_en
+Line 6224 
 codestart = cd->name_table + re->name_en
  cd->start_code = codestart;
  cd->hwm = cworkspace;
  cd->req_varyopt = 0;
- cd->nopartial = FALSE;
  cd->had_accept = FALSE;
  /* Set up a starting, non-extracting bracket, then compile the expression. On
  error, errorcode will be set non-zero, so we don't need to look at the result
  of the function here. */
- ptr = (const uschar *)pattern;
+ ptr = (const uschar *)pattern + skipatstart;
  code = (uschar *)codestart;
  *code = OP_BRA;
  (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
    &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
  re->top_bracket = cd->bracount;
  re->top_backref = cd->top_backref;
+ re->flags = cd->external_flags;
- if (cd->nopartial) re->options |= PCRE_NOPARTIAL;
  if (cd->had_accept) reqbyte = -1;   /* Must disable after (*ACCEPT) */
  /* If not reached end of pattern on success, there's an excess bracket. */
-Line 5828 
 if (errorcode != 0)
+Line 6281 
 if (errorcode != 0)
    PCRE_EARLY_ERROR_RETURN:
    *erroroffset = ptr - (const uschar *)pattern;
    PCRE_EARLY_ERROR_RETURN2:
-   *errorptr = error_texts[errorcode];
+   *errorptr = find_error_text(errorcode);
    if (errorcodeptr != NULL) *errorcodeptr = errorcode;
    return NULL;
    }
-Line 5857 
 if ((re->options & PCRE_ANCHORED) == 0)
+Line 6310 
 if ((re->options & PCRE_ANCHORED) == 0)
        int ch = firstbyte & 255;
        re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
           cd->fcc[ch] == ch)? ch : firstbyte;
-       re->options |= PCRE_FIRSTSET;
+       re->flags |= PCRE_FIRSTSET;
        }
      else if (is_startline(codestart, 0, cd->backref_map))
-       re->options |= PCRE_STARTLINE;
+       re->flags |= PCRE_STARTLINE;
      }
    }
-Line 5874 
 if (reqbyte >= 0 &&
+Line 6327 
 if (reqbyte >= 0 &&
    int ch = reqbyte & 255;
    re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
      cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
-   re->options |= PCRE_REQCHSET;
+   re->flags |= PCRE_REQCHSET;
    }
  /* Print out the compiled data if debugging is enabled. This is never the
-Line 5885 
 case when building a production library.
+Line 6338 
 case when building a production library.
  printf("Length = %d top_bracket = %d top_backref = %d\n",
    length, re->top_bracket, re->top_backref);
- if (re->options != 0)
+ printf("Options=%08x\n", re->options);
-   {
-   printf("%s%s%s%s%s%s%s%s%s\n",
-     ((re->options & PCRE_NOPARTIAL) != 0)? "nopartial " : "",
-     ((re->options & PCRE_ANCHORED) != 0)? "anchored " : "",
-     ((re->options & PCRE_CASELESS) != 0)? "caseless " : "",
-     ((re->options & PCRE_EXTENDED) != 0)? "extended " : "",
-     ((re->options & PCRE_MULTILINE) != 0)? "multiline " : "",
-     ((re->options & PCRE_DOTALL) != 0)? "dotall " : "",
-     ((re->options & PCRE_DOLLAR_ENDONLY) != 0)? "endonly " : "",
-     ((re->options & PCRE_EXTRA) != 0)? "extra " : "",
-     ((re->options & PCRE_UNGREEDY) != 0)? "ungreedy " : "");
-   }
- if ((re->options & PCRE_FIRSTSET) != 0)
+ if ((re->flags & PCRE_FIRSTSET) != 0)
    {
    int ch = re->first_byte & 255;
    const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
-Line 5908 
 if ((re->options & PCRE_FIRSTSET) != 0)
+Line 6349 
 if ((re->options & PCRE_FIRSTSET) != 0)
      else printf("First char = \\x%02x%s\n", ch, caseless);
    }
- if ((re->options & PCRE_REQCHSET) != 0)
+ if ((re->flags & PCRE_REQCHSET) != 0)
    {
    int ch = re->req_byte & 255;
    const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
-Line 5925 
 was compiled can be seen. */
+Line 6366 
 was compiled can be seen. */
  if (code - codestart > length)
    {
    (pcre_free)(re);
-   *errorptr = error_texts[ERR23];
+   *errorptr = find_error_text(ERR23);
    *erroroffset = ptr - (uschar *)pattern;
    if (errorcodeptr != NULL) *errorcodeptr = ERR23;
    return NULL;

 Legend:



Removed from v.210
 


changed lines


 
Added in v.345
 Legend:



Removed from v.210
 


changed lines


 
Added in v.345
-Removed from v.210
+Added in v.345

	ViewVC Help
Powered by ViewVC 1.1.5