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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 550 by ph10,
Sun Oct 10 16:24:11 2010 UTC
+revision 773 by ph10,
Wed Nov 30 18:10:27 2011 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2010 University of Cambridge
+            Copyright (c) 1997-2011 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 88 
 so this number is very generous.
  The same workspace is used during the second, actual compile phase for
  remembering forward references to groups so that they can be filled in at the
  end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
- is 4 there is plenty of room. */
+ is 4 there is plenty of room for most patterns. However, the memory can get
+ filled up by repetitions of forward references, for example patterns like
+ /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
+ that the workspace is expanded using malloc() in this situation. The value
+ below is therefore a minimum, and we put a maximum on it for safety. The
+ minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
+ kicks in at the same number of forward references in all cases. */
- #define COMPILE_WORK_SIZE (4096)
+ #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_CHECK (COMPILE_WORK_SIZE - 100)
+ #define WORK_SIZE_SAFETY_MARGIN (100)
  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
-Line 393 
 static const char error_texts[] =
+Line 400 
 static const char error_texts[] =
    "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"
+   "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"
-Line 408 
 static const char error_texts[] =
+Line 415 
 static const char error_texts[] =
    "different names for subpatterns of the same number are not allowed\0"
    "(*MARK) must have an argument\0"
    "this version of PCRE is not compiled with PCRE_UCP support\0"
+   "\\c must be followed by an ASCII character\0"
+   "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
+   /* 70 */
+   "internal error: unknown opcode in find_fixedlength()\0"
+   "\\N is not supported in a class\0"
+   "too many forward references\0"
    ;
  /* Table to identify digits and hex digits. This is used when compiling
-Line 544 
 static const unsigned char ebcdic_charta
+Line 557 
 static const unsigned char ebcdic_charta
  /* Definition to allow mutual recursion */
  static BOOL
-   compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
+   compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
      int *, int *, branch_chain *, compile_data *, int *);
-Line 576 
 return s;
+Line 589 
 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)
+ {
+ uschar *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 = (pcre_malloc)(newsize);
+ if (newspace == NULL) return ERR21;
+ memcpy(newspace, cd->start_workspace, cd->workspace_size);
+ cd->hwm = (uschar *)newspace + (cd->hwm - cd->start_workspace);
+ if (cd->workspace_size > COMPILE_WORK_SIZE)
+   (pcre_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 uschar *p)
+ {
+ if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
+ while ((digitab[*p] & ctype_digit) != 0) p++;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if (*p++ != CHAR_COMMA) return FALSE;
+ if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
+ if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
+ while ((digitab[*p] & ctype_digit) != 0) p++;
+ return (*p == CHAR_RIGHT_CURLY_BRACKET);
+ }
+ /*************************************************
  *            Handle escapes                      *
  *************************************************/
-Line 641 
 else
+Line 725 
 else
      case CHAR_l:
      case CHAR_L:
+     *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 ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
+            && (digitab[ptr[3]] & ctype_xdigit) != 0 && (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:
-     *errorcodeptr = ERR37;
+     /* In JavaScript, \U is an uppercase U letter. */
+     if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by one of a number of specific things:
+     /* 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
-Line 663 
 else
+Line 777 
 else
      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;
-Line 791 
 else
+Line 906 
 else
      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 ((digitab[ptr[1]] & ctype_xdigit) != 0 && (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;
+       }
      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
        {
        const uschar *pt = ptr + 2;
-Line 841 
 else
+Line 978 
 else
      break;
      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
-     This coding is ASCII-specific, but then the whole concept of \cx is
+     An error is given if the byte following \c is not an ASCII character. This
+     coding is ASCII-specific, but then the whole concept of \cx is
      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
      case CHAR_c:
-Line 851 
 else
+Line 989 
 else
        *errorcodeptr = ERR2;
        break;
        }
+ #ifndef EBCDIC    /* ASCII/UTF-8 coding */
- #ifndef EBCDIC  /* ASCII/UTF-8 coding */
+     if (c > 127)  /* Excludes all non-ASCII in either mode */
+       {
+       *errorcodeptr = ERR68;
+       break;
+       }
      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
      c ^= 0x40;
- #else           /* EBCDIC coding */
+ #else             /* EBCDIC coding */
      if (c >= CHAR_a && c <= CHAR_z) c += 64;
      c ^= 0xC0;
  #endif
-Line 879 
 else
+Line 1021 
 else
    }
  /* Perl supports \N{name} for character names, as well as plain \N for "not
- newline". PCRE does not support \N{name}. */
+ 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)
+ 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. */
-Line 991 
 return -1;
+Line 1135 
 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 == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
- if (*p++ != CHAR_COMMA) return FALSE;
- if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
- if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
- while ((digitab[*p] & ctype_digit) != 0) p++;
- return (*p == CHAR_RIGHT_CURLY_BRACKET);
- }
- /*************************************************
  *         Read repeat counts                     *
  *************************************************/
-Line 1099 
 top-level call starts at the beginning o
+Line 1210 
 top-level call starts at the beginning o
  start at a parenthesis. It scans along a pattern's text looking for capturing
  subpatterns, and counting them. If it finds a named pattern that matches the
  name it is given, it returns its number. Alternatively, if the name is NULL, it
- returns when it reaches a given numbered subpattern. We know that if (?P< is
+ returns when it reaches a given numbered subpattern. Recursion is used to keep
- encountered, the name will be terminated by '>' because that is checked in the
+ track of subpatterns that reset the capturing group numbers - the (?| feature.
- first pass. 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:
    ptrptr       address of the current character pointer (updated)
-Line 1110 
 Arguments:
+Line 1232 
 Arguments:
    name         name to seek, or NULL if seeking a numbered subpattern
    lorn         name length, or subpattern number if name is NULL
    xmode        TRUE if we are in /x mode
+   utf8         TRUE if we are in UTF-8 mode
    count        pointer to the current capturing subpattern number (updated)
  Returns:       the number of the named subpattern, or -1 if not found
-Line 1117 
 Returns:       the number of the named s
+Line 1240 
 Returns:       the number of the named s
  static int
  find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
-   BOOL xmode, int *count)
+   BOOL xmode, BOOL utf8, int *count)
  {
  uschar *ptr = *ptrptr;
  int start_count = *count;
-Line 1202 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1325 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    }
  /* Past any initial parenthesis handling, scan for parentheses or vertical
- bars. */
+ 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 != 0; ptr++)
+ for (; ptr < cd->end_pattern; ptr++)
    {
    /* Skip over backslashed characters and also entire \Q...\E */
-Line 1278 
 for (; *ptr != 0; ptr++)
+Line 1403 
 for (; *ptr != 0; ptr++)
    if (xmode && *ptr == CHAR_NUMBER_SIGN)
      {
-     while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
+     ptr++;
+     while (*ptr != 0)
+       {
+       if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+       ptr++;
+ #ifdef SUPPORT_UTF8
+       if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+       }
      if (*ptr == 0) goto FAIL_EXIT;
      continue;
      }
-Line 1287 
 for (; *ptr != 0; ptr++)
+Line 1420 
 for (; *ptr != 0; ptr++)
    if (*ptr == CHAR_LEFT_PARENTHESIS)
      {
-     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
+     int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
      if (rc > 0) return rc;
      if (*ptr == 0) goto FAIL_EXIT;
      }
-Line 1333 
 Arguments:
+Line 1466 
 Arguments:
    name         name to seek, or NULL if seeking a numbered subpattern
    lorn         name length, or subpattern number if name is NULL
    xmode        TRUE if we are in /x mode
+   utf8         TRUE if we are in UTF-8 mode
  Returns:       the number of the found subpattern, or -1 if not found
  */
  static int
- find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
+ find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
+   BOOL utf8)
  {
  uschar *ptr = (uschar *)cd->start_pattern;
  int count = 0;
-Line 1351 
 matching closing parens. That is why we
+Line 1486 
 matching closing parens. That is why we
  for (;;)
    {
-   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
+   rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
    if (rc > 0 || *ptr++ == 0) break;
    }
-Line 1367 
 return rc;
+Line 1502 
 return rc;
  /* This is called by several functions that scan a compiled expression looking
  for a fixed first character, or an anchoring op code etc. It skips over things
- that do not influence this. For some calls, a change of option is important.
+ that do not influence this. For some calls, it makes sense to skip negative
- For some calls, it makes sense to skip negative forward and all backward
+ forward and all backward assertions, and also the \b assertion; for others it
- assertions, and also the \b assertion; for others it does not.
+ does not.
  Arguments:
    code         pointer to the start of the group
-   options      pointer to external options
-   optbit       the option bit whose changing is significant, or
-                  zero if none are
    skipassert   TRUE if certain assertions are to be skipped
  Returns:       pointer to the first significant opcode
  */
  static const uschar*
- first_significant_code(const uschar *code, int *options, int optbit,
+ first_significant_code(const uschar *code, BOOL skipassert)
-   BOOL skipassert)
  {
  for (;;)
    {
    switch ((int)*code)
      {
-     case OP_OPT:
-     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
-       *options = (int)code[1];
-     code += 2;
-     break;
      case OP_ASSERT_NOT:
      case OP_ASSERTBACK:
      case OP_ASSERTBACK_NOT:
-Line 1444 
 and doing the check at the end; a flag s
+Line 1569 
 and doing the check at the end; a flag s
  Arguments:
    code     points to the start of the pattern (the bracket)
-   options  the compiling options
+   utf8     TRUE in UTF-8 mode
    atend    TRUE if called when the pattern is complete
    cd       the "compile data" structure
  Returns:   the fixed length,
               or -1 if there is no fixed length,
-              or -2 if \C was encountered
+              or -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, BOOL atend, compile_data *cd)
+ find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
  {
  int length = -1;
-Line 1472 
 for (;;)
+Line 1598 
 for (;;)
    register int op = *cc;
    switch (op)
      {
+     /* We only need to continue for OP_CBRA (normal capturing bracket) and
+     OP_BRA (normal non-capturing bracket) because the other variants of these
+     opcodes are all concerned with unlimited repeated groups, which of course
+     are not of fixed length. */
      case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
+     case OP_ONCE_NC:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
+     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      do cc += GET(cc, 1); while (*cc == OP_ALT);
      cc += 1 + LINK_SIZE;
      break;
-     /* Reached end of a branch; if it's a ket it is the end of a nested
+     /* 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 1508 
 for (;;)
+Line 1642 
 for (;;)
      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
      if (cc > cs && cc < ce) return -1;                /* Recursion */
-     d = find_fixedlength(cs + 2, options, atend, cd);
+     d = find_fixedlength(cs + 2, utf8, atend, cd);
      if (d < 0) return d;
      branchlength += d;
      cc += 1 + LINK_SIZE;
-Line 1525 
 for (;;)
+Line 1659 
 for (;;)
      /* Skip over things that don't match chars */
-     case OP_REVERSE:
+     case OP_MARK:
+     case OP_PRUNE_ARG:
+     case OP_SKIP_ARG:
+     case OP_THEN_ARG:
+     cc += cc[1] + _pcre_OP_lengths[*cc];
+     break;
+     case OP_CALLOUT:
+     case OP_CIRC:
+     case OP_CIRCM:
+     case OP_CLOSE:
+     case OP_COMMIT:
      case OP_CREF:
-     case OP_NCREF:
-     case OP_RREF:
-     case OP_NRREF:
      case OP_DEF:
-     case OP_OPT:
+     case OP_DOLL:
-     case OP_CALLOUT:
+     case OP_DOLLM:
-     case OP_SOD:
-     case OP_SOM:
-     case OP_SET_SOM:
      case OP_EOD:
      case OP_EODN:
-     case OP_CIRC:
+     case OP_FAIL:
-     case OP_DOLL:
+     case OP_NCREF:
+     case OP_NRREF:
      case OP_NOT_WORD_BOUNDARY:
+     case OP_PRUNE:
+     case OP_REVERSE:
+     case OP_RREF:
+     case OP_SET_SOM:
+     case OP_SKIP:
+     case OP_SOD:
+     case OP_SOM:
+     case OP_THEN:
      case OP_WORD_BOUNDARY:
      cc += _pcre_OP_lengths[*cc];
      break;
-Line 1548 
 for (;;)
+Line 1696 
 for (;;)
      /* Handle literal characters */
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      branchlength++;
      cc += 2;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
+     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
  #endif
      break;
-Line 1562 
 for (;;)
+Line 1710 
 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;
  #ifdef SUPPORT_UTF8
-     if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
+     if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
-       cc += _pcre_utf8_table4[cc[-1] & 0x3f];
  #endif
      break;
-Line 1583 
 for (;;)
+Line 1733 
 for (;;)
      cc += 2;
      /* Fall through */
+     case OP_HSPACE:
+     case OP_VSPACE:
+     case OP_NOT_HSPACE:
+     case OP_NOT_VSPACE:
      case OP_NOT_DIGIT:
      case OP_DIGIT:
      case OP_NOT_WHITESPACE:
-Line 1595 
 for (;;)
+Line 1749 
 for (;;)
      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;
-Line 1614 
 for (;;)
+Line 1769 
 for (;;)
      switch (*cc)
        {
+       case OP_CRPLUS:
+       case OP_CRMINPLUS:
        case OP_CRSTAR:
        case OP_CRMINSTAR:
        case OP_CRQUERY:
-Line 1634 
 for (;;)
+Line 1791 
 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 1668 
 _pcre_find_bracket(const uschar *code, B
+Line 1908 
 _pcre_find_bracket(const uschar *code, B
  for (;;)
    {
    register int c = *code;
    if (c == OP_END) return NULL;
    /* XCLASS is used for classes that cannot be represented just by a bit
-Line 1686 
 for (;;)
+Line 1927 
 for (;;)
    /* Handle capturing bracket */
-   else if (c == OP_CBRA)
+   else if (c == OP_CBRA || c == OP_SCBRA ||
+            c == OP_CBRAPOS || c == OP_SCBRAPOS)
      {
      int n = GET2(code, 1+LINK_SIZE);
      if (n == number) return (uschar *)code;
-Line 1728 
 for (;;)
+Line 1970 
 for (;;)
        break;
        case OP_THEN_ARG:
-       code += code[1+LINK_SIZE];
+       code += code[1];
        break;
        }
-Line 1744 
 for (;;)
+Line 1986 
 for (;;)
      if (utf8) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
        case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
        case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
        case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
+       case OP_MINQUERYI:
        case OP_POSQUERY:
+       case OP_POSQUERYI:
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
-Line 1834 
 for (;;)
+Line 2089 
 for (;;)
        break;
        case OP_THEN_ARG:
-       code += code[1+LINK_SIZE];
+       code += code[1];
        break;
        }
-Line 1850 
 for (;;)
+Line 2105 
 for (;;)
      if (utf8) switch(c)
        {
        case OP_CHAR:
-       case OP_CHARNC:
+       case OP_CHARI:
        case OP_EXACT:
+       case OP_EXACTI:
        case OP_UPTO:
+       case OP_UPTOI:
        case OP_MINUPTO:
+       case OP_MINUPTOI:
        case OP_POSUPTO:
+       case OP_POSUPTOI:
        case OP_STAR:
+       case OP_STARI:
        case OP_MINSTAR:
+       case OP_MINSTARI:
        case OP_POSSTAR:
+       case OP_POSSTARI:
        case OP_PLUS:
+       case OP_PLUSI:
        case OP_MINPLUS:
+       case OP_MINPLUSI:
        case OP_POSPLUS:
+       case OP_POSPLUSI:
        case OP_QUERY:
+       case OP_QUERYI:
        case OP_MINQUERY:
+       case OP_MINQUERYI:
        case OP_POSQUERY:
+       case OP_POSQUERYI:
        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
        break;
        }
-Line 1902 
 could_be_empty_branch(const uschar *code
+Line 2170 
 could_be_empty_branch(const uschar *code
    compile_data *cd)
  {
  register int c;
- for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
+ for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
       code < endcode;
-      code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
+      code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
    {
    const uschar *ccode;
-Line 1920 
 for (code = first_significant_code(code
+Line 2188 
 for (code = first_significant_code(code
      continue;
      }
-   /* Groups with zero repeats can of course be empty; skip them. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
-     {
-     code += _pcre_OP_lengths[c];
-     do code += GET(code, 1); while (*code == OP_ALT);
-     c = *code;
-     continue;
-     }
    /* For a recursion/subroutine call, if its end has been reached, which
-   implies a subroutine call, we can scan it. */
+   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)
      {
-     BOOL empty_branch = FALSE;
+     const uschar *scode;
-     const uschar *scode = cd->start_code + GET(code, 1);
+     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 */
      do
        {
        if (could_be_empty_branch(scode, endcode, utf8, cd))
-Line 1948 
 for (code = first_significant_code(code
+Line 2223 
 for (code = first_significant_code(code
        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 += _pcre_OP_lengths[c];
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
+   /* A nested group that is already marked as "could be empty" can just be
+   skipped. */
+   if (c == OP_SBRA  || c == OP_SBRAPOS ||
+       c == OP_SCBRA || c == OP_SCBRAPOS)
+     {
+     do code += GET(code, 1); while (*code == OP_ALT);
+     c = *code;
+     continue;
+     }
    /* For other groups, scan the branches. */
-   if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
+   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 */
-Line 2041 
 for (code = first_significant_code(code
+Line 2342 
 for (code = first_significant_code(code
      case OP_ALLANY:
      case OP_ANYBYTE:
      case OP_CHAR:
-     case OP_CHARNC:
+     case OP_CHARI:
      case OP_NOT:
+     case OP_NOTI:
      case OP_PLUS:
      case OP_MINPLUS:
      case OP_POSPLUS:
-Line 2082 
 for (code = first_significant_code(code
+Line 2384 
 for (code = first_significant_code(code
      case OP_KET:
      case OP_KETRMAX:
      case OP_KETRMIN:
+     case OP_KETRPOS:
      case OP_ALT:
      return TRUE;
-Line 2090 
 for (code = first_significant_code(code
+Line 2393 
 for (code = first_significant_code(code
  #ifdef SUPPORT_UTF8
      case OP_STAR:
+     case OP_STARI:
      case OP_MINSTAR:
+     case OP_MINSTARI:
      case OP_POSSTAR:
+     case OP_POSSTARI:
      case OP_QUERY:
+     case OP_QUERYI:
      case OP_MINQUERY:
+     case OP_MINQUERYI:
      case OP_POSQUERY:
+     case OP_POSQUERYI:
      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
      break;
      case OP_UPTO:
+     case OP_UPTOI:
      case OP_MINUPTO:
+     case OP_MINUPTOI:
      case OP_POSUPTO:
+     case OP_POSUPTOI:
      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
      break;
  #endif
-Line 2115 
 for (code = first_significant_code(code
+Line 2427 
 for (code = first_significant_code(code
      break;
      case OP_THEN_ARG:
-     code += code[1+LINK_SIZE];
+     code += code[1];
      break;
      /* None of the remaining opcodes are required to match a character. */
-Line 2138 
 return TRUE;
+Line 2450 
 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
-Line 2188 
 where Perl recognizes it as the POSIX cl
+Line 2502 
 where Perl recognizes it as the POSIX cl
  "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.
  Arguments:
    ptr      pointer to the initial [
    endptr   where to return the end pointer
-Line 2202 
 int terminator;          /* Don't combin
+Line 2527 
 int terminator;          /* Don't combin
  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
  for (++ptr; *ptr != 0; ptr++)
    {
-   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
+   if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
+     ptr++;
+   else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
+   else
      {
-     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
      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 2515 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2847 
 if ((options & PCRE_EXTENDED) != 0)
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
      if (*ptr == CHAR_NUMBER_SIGN)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+         }
        }
      else break;
      }
-Line 2552 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2891 
 if ((options & PCRE_EXTENDED) != 0)
      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
      if (*ptr == CHAR_NUMBER_SIGN)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
+         {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
+         }
        }
      else break;
      }
-Line 2578 
 if (next >= 0) switch(op_code)
+Line 2924 
 if (next >= 0) switch(op_code)
  #endif
    return c != next;
-   /* For CHARNC (caseless character) we must check the other case. If we have
+   /* 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_CHARNC:
+   case OP_CHARI:
  #ifdef SUPPORT_UTF8
    GETCHARTEST(c, previous);
  #else
-Line 2605 
 if (next >= 0) switch(op_code)
+Line 2951 
 if (next >= 0) switch(op_code)
  #endif  /* SUPPORT_UTF8 */
    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
-   /* For OP_NOT, its data is always a single-byte character. */
+   /* 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;
-   if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
    if (utf8)
      {
-Line 2714 
 replaced by OP_PROP codes when PCRE_UCP
+Line 3064 
 replaced by OP_PROP codes when PCRE_UCP
  switch(op_code)
    {
    case OP_CHAR:
-   case OP_CHARNC:
+   case OP_CHARI:
  #ifdef SUPPORT_UTF8
    GETCHARTEST(c, previous);
  #else
-Line 2901 
 Arguments:
+Line 3251 
 Arguments:
    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
    reqbyteptr     set to the last literal character required, else < 0
    bcptr          points to current branch chain
+   cond_depth     conditional nesting depth
    cd             contains pointers to tables etc.
    lengthptr      NULL during the real compile phase
                   points to length accumulator during pre-compile phase
-Line 2912 
 Returns:         TRUE on success
+Line 3263 
 Returns:         TRUE on success
  static BOOL
  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
-   compile_data *cd, int *lengthptr)
+   int cond_depth, compile_data *cd, int *lengthptr)
  {
  int repeat_type, op_type;
  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
-Line 2921 
 int greedy_default, greedy_non_default;
+Line 3272 
 int greedy_default, greedy_non_default;
  int firstbyte, reqbyte;
  int zeroreqbyte, zerofirstbyte;
  int req_caseopt, reqvary, tempreqvary;
- int options = *optionsptr;
+ int options = *optionsptr;               /* May change dynamically */
  int after_manual_callout = 0;
  int length_prevgroup = 0;
  register int c;
-Line 2939 
 uschar *previous_callout = NULL;
+Line 3290 
 uschar *previous_callout = NULL;
  uschar *save_hwm = NULL;
  uschar classbits[32];
+ /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
+ must not do this for other options (e.g. PCRE_EXTENDED) because they may change
+ dynamically as we process the pattern. */
  #ifdef SUPPORT_UTF8
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
-Line 2947 
 uschar *class_utf8data_base;
+Line 3302 
 uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
- uschar *utf8_char = NULL;
  #endif
  #ifdef PCRE_DEBUG
-Line 2998 
 for (;; ptr++)
+Line 3352 
 for (;; ptr++)
    int subfirstbyte;
    int terminator;
    int mclength;
+   int tempbracount;
    uschar mcbuffer[8];
    /* Get next byte in the pattern */
-Line 3022 
 for (;; ptr++)
+Line 3377 
 for (;; ptr++)
  #ifdef PCRE_DEBUG
      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
  #endif
-     if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
+     if (code > cd->start_workspace + cd->workspace_size -
+         WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
        {
        *errorcodeptr = ERR52;
        goto FAILED;
-Line 3045 
 for (;; ptr++)
+Line 3401 
 for (;; ptr++)
        }
      *lengthptr += (int)(code - last_code);
-     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
+     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
+       c));
      /* If "previous" is set and it is not at the start of the work space, move
      it back to there, in order to avoid filling up the work space. Otherwise,
-Line 3071 
 for (;; ptr++)
+Line 3428 
 for (;; ptr++)
    /* In the real compile phase, just check the workspace used by the forward
    reference list. */
-   else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
+   else if (cd->hwm > cd->start_workspace + cd->workspace_size -
+            WORK_SIZE_SAFETY_MARGIN)
      {
      *errorcodeptr = ERR52;
      goto FAILED;
-Line 3119 
 for (;; ptr++)
+Line 3477 
 for (;; ptr++)
      previous_callout = NULL;
      }
-   /* In extended mode, skip white space and comments */
+   /* In extended mode, skip white space and comments. */
    if ((options & PCRE_EXTENDED) != 0)
      {
      if ((cd->ctypes[c] & ctype_space) != 0) continue;
      if (c == CHAR_NUMBER_SIGN)
        {
-       while (*(++ptr) != 0)
+       ptr++;
+       while (*ptr != 0)
          {
          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
+         ptr++;
+ #ifdef SUPPORT_UTF8
+         if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
+ #endif
          }
        if (*ptr != 0) continue;
-Line 3173 
 for (;; ptr++)
+Line 3536 
 for (;; ptr++)
      the setting of any following char as a first character. */
      case CHAR_CIRCUMFLEX_ACCENT:
+     previous = NULL;
      if ((options & PCRE_MULTILINE) != 0)
        {
        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
+       *code++ = OP_CIRCM;
        }
-     previous = NULL;
+     else *code++ = OP_CIRC;
-     *code++ = OP_CIRC;
      break;
      case CHAR_DOLLAR_SIGN:
      previous = NULL;
-     *code++ = OP_DOLL;
+     *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
      break;
      /* There can never be a first char if '.' is first, whatever happens about
-Line 3455 
 for (;; ptr++)
+Line 3819 
 for (;; ptr++)
          if (*errorcodeptr != 0) goto FAILED;
          if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
+         else if (-c == ESC_N)            /* \N is not supported in a class */
+           {
+           *errorcodeptr = ERR71;
+           goto FAILED;
+           }
          else if (-c == ESC_Q)            /* Handle start of quoted string */
            {
            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
-Line 3503 
 for (;; ptr++)
+Line 3872 
 for (;; ptr++)
              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
              continue;
+             /* Perl 5.004 onwards omits VT from \s, but we must preserve it
+             if it was previously set by something earlier in the character
+             class. */
              case ESC_s:
-             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
+             classbits[0] |= cbits[cbit_space];
-             classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */
+             classbits[1] |= cbits[cbit_space+1] & ~0x08;
+             for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
              continue;
              case ESC_S:
-Line 3924 
 for (;; ptr++)
+Line 4298 
 for (;; ptr++)
      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
+     operate on single-bytes characters only. This is an historical hangover.
-     we can tidy these opcodes to handle multi-byte characters.
+     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
+-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
-     that OP_NOT does not support multibyte characters. In the positive case, it
+     Note that OP_NOT[I] does not support multibyte characters. In the positive
-     can cause firstbyte to be set. Otherwise, there can be no first char if
+     case, it can cause firstbyte to be set. Otherwise, there can be no first
-     this item is first, whatever repeat count may follow. In the case of
+     char if this item is first, whatever repeat count may follow. In the case
-     reqbyte, save the previous value for reinstating. */
+     of reqbyte, save the previous value for reinstating. */
  #ifdef SUPPORT_UTF8
      if (class_charcount == 1 && !class_utf8 &&
-Line 3943 
 for (;; ptr++)
+Line 4317 
 for (;; ptr++)
        {
        zeroreqbyte = reqbyte;
-       /* The OP_NOT opcode works on one-byte characters only. */
+       /* The OP_NOT[I] opcodes work on one-byte characters only. */
        if (negate_class)
          {
          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
          zerofirstbyte = firstbyte;
-         *code++ = OP_NOT;
+         *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
          *code++ = class_lastchar;
          break;
          }
-Line 4077 
 for (;; ptr++)
+Line 4451 
 for (;; ptr++)
      op_type = 0;                    /* Default single-char op codes */
      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
-     /* Save start of previous item, in case we have to move it up to make space
+     /* Save start of previous item, in case we have to move it up in order to
-     for an inserted OP_ONCE for the additional '+' extension. */
+     insert something before it. */
      tempcode = previous;
-Line 4101 
 for (;; ptr++)
+Line 4475 
 for (;; ptr++)
        }
      else repeat_type = greedy_default;
+     /* If previous was a recursion call, wrap it in atomic brackets so that
+     previous becomes the atomic group. All recursions were so wrapped in the
+     past, but it no longer happens for non-repeated recursions. In fact, the
+     repeated ones could be re-implemented independently so as not to need this,
+     but for the moment we rely on the code for repeating groups. */
+     if (*previous == OP_RECURSE)
+       {
+       memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
+       *previous = OP_ONCE;
+       PUT(previous, 1, 2 + 2*LINK_SIZE);
+       previous[2 + 2*LINK_SIZE] = OP_KET;
+       PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
+       code += 2 + 2 * LINK_SIZE;
+       length_prevgroup = 3 + 3*LINK_SIZE;
+       /* When actually compiling, we need to check whether this was a forward
+       reference, and if so, adjust the offset. */
+       if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
+         {
+         int offset = GET(cd->hwm, -LINK_SIZE);
+         if (offset == previous + 1 - cd->start_code)
+           PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
+         }
+       }
+     /* Now handle repetition for the different types of item. */
      /* If previous was a character match, abolish the item and generate a
      repeat item instead. If a char item has a minumum of more than one, ensure
      that it is set in reqbyte - it might not be if a sequence such as x{3} is
      the first thing in a branch because the x will have gone into firstbyte
      instead.  */
-     if (*previous == OP_CHAR || *previous == OP_CHARNC)
+     if (*previous == OP_CHAR || *previous == OP_CHARI)
        {
+       op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
        /* Deal with UTF-8 characters that take up more than one byte. It's
        easier to write this out separately than try to macrify it. Use c to
        hold the length of the character in bytes, plus 0x80 to flag that it's a
-Line 4153 
 for (;; ptr++)
+Line 4558 
 for (;; ptr++)
      /* If previous was a single negated character ([^a] or similar), we use
      one of the special opcodes, replacing it. The code is shared with single-
      character repeats by setting opt_type to add a suitable offset into
-     repeat_type. We can also test for auto-possessification. OP_NOT is
+     repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI
-     currently used only for single-byte chars. */
+     are currently used only for single-byte chars. */
-     else if (*previous == OP_NOT)
+     else if (*previous == OP_NOT || *previous == OP_NOTI)
        {
-       op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */
+       op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
        c = previous[1];
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-Line 4355 
 for (;; ptr++)
+Line 4760 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
               *previous == OP_XCLASS ||
  #endif
-              *previous == OP_REF)
+              *previous == OP_REF ||
+              *previous == OP_REFI)
        {
        if (repeat_max == 0)
          {
-Line 4389 
 for (;; ptr++)
+Line 4795 
 for (;; ptr++)
        }
      /* If previous was a bracket group, we may have to replicate it in certain
-     cases. */
+     cases. Note that at this point we can encounter only the "basic" bracket
+     opcodes such as BRA and CBRA, as this is the place where they get converted
+     into the more special varieties such as BRAPOS and SBRA. A test for >=
+     OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
+     ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
+     repetition of assertions, but now it does, for Perl compatibility. */
-     else if (*previous == OP_BRA  || *previous == OP_CBRA ||
+     else if (*previous >= OP_ASSERT && *previous <= OP_COND)
-              *previous == OP_ONCE || *previous == OP_COND)
        {
        register int i;
-       int ketoffset = 0;
        int len = (int)(code - previous);
        uschar *bralink = NULL;
+       uschar *brazeroptr = NULL;
-       /* Repeating a DEFINE group is pointless */
+       /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
+       we just ignore the repeat. */
        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
-         {
+         goto END_REPEAT;
-         *errorcodeptr = ERR55;
-         goto FAILED;
+       /* There is no sense in actually repeating assertions. The only potential
-         }
+       use of repetition is in cases when the assertion is optional. Therefore,
+       if the minimum is greater than zero, just ignore the repeat. If the
+       maximum is not not zero or one, set it to 1. */
-       /* If the maximum repeat count is unlimited, find the end of the bracket
+       if (*previous < OP_ONCE)    /* Assertion */
-       by scanning through from the start, and compute the offset back to it
+         {
-       from the current code pointer. There may be an OP_OPT setting following
+         if (repeat_min > 0) goto END_REPEAT;
-       the final KET, so we can't find the end just by going back from the code
+         if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
-       pointer. */
-       if (repeat_max == -1)
-         {
-         register uschar *ket = previous;
-         do ket += GET(ket, 1); while (*ket != OP_KET);
-         ketoffset = (int)(code - ket);
          }
        /* The case of a zero minimum is special because of the need to stick
-Line 4438 
 for (;; ptr++)
+Line 4844 
 for (;; ptr++)
          **   goto END_REPEAT;
          **   }
-         However, that fails when a group is referenced as a subroutine from
+         However, that fails when a group or a subgroup within it is referenced
-         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
+         as a subroutine from elsewhere in the pattern, so now we stick in
-         so that it is skipped on execution. As we don't have a list of which
+         OP_SKIPZERO in front of it so that it is skipped on execution. As we
-         groups are referenced, we cannot do this selectively.
+         don't have a list of which groups are referenced, we cannot do this
+         selectively.
          If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
          and do no more at this point. However, we do need to adjust any
-Line 4461 
 for (;; ptr++)
+Line 4868 
 for (;; ptr++)
              *previous++ = OP_SKIPZERO;
              goto END_REPEAT;
              }
+           brazeroptr = previous;    /* Save for possessive optimizing */
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 4523 
 for (;; ptr++)
+Line 4931 
 for (;; ptr++)
              *lengthptr += delta;
              }
-           /* This is compiling for real */
+           /* This is compiling for real. If there is a set first byte for
+           the group, and we have not yet set a "required byte", set it. Make
+           sure there is enough workspace for copying forward references before
+           doing the copy. */
            else
              {
              if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
              for (i = 1; i < repeat_min; i++)
                {
                uschar *hc;
                uschar *this_hwm = cd->hwm;
                memcpy(code, previous, len);
+               while (cd->hwm > cd->start_workspace + cd->workspace_size -
+                      WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
+                 {
+                 int save_offset = save_hwm - cd->start_workspace;
+                 int this_offset = this_hwm - cd->start_workspace;
+                 *errorcodeptr = expand_workspace(cd);
+                 if (*errorcodeptr != 0) goto FAILED;
+                 save_hwm = (uschar *)cd->start_workspace + save_offset;
+                 this_hwm = (uschar *)cd->start_workspace + this_offset;
+                 }
                for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
                  {
                  PUT(cd->hwm, 0, GET(hc, 0) + len);
-Line 4562 
 for (;; ptr++)
+Line 4986 
 for (;; ptr++)
          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
          a 64-bit integer type when available, otherwise double. */
          if (lengthptr != NULL && repeat_max > 0)
            {
            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
-Line 4600 
 for (;; ptr++)
+Line 5024 
 for (;; ptr++)
              }
            memcpy(code, previous, len);
+           /* Ensure there is enough workspace for forward references before
+           copying them. */
+           while (cd->hwm > cd->start_workspace + cd->workspace_size -
+                  WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
+             {
+             int save_offset = save_hwm - cd->start_workspace;
+             int this_offset = this_hwm - cd->start_workspace;
+             *errorcodeptr = expand_workspace(cd);
+             if (*errorcodeptr != 0) goto FAILED;
+             save_hwm = (uschar *)cd->start_workspace + save_offset;
+             this_hwm = (uschar *)cd->start_workspace + this_offset;
+             }
            for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
              {
              PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
-Line 4625 
 for (;; ptr++)
+Line 5064 
 for (;; ptr++)
            }
          }
-       /* If the maximum is unlimited, set a repeater in the final copy. We
+       /* If the maximum is unlimited, set a repeater in the final copy. For
-       can't just offset backwards from the current code point, because we
+       ONCE brackets, that's all we need to do. However, possessively repeated
-       don't know if there's been an options resetting after the ket. The
+       ONCE brackets can be converted into non-capturing brackets, as the
-       correct offset was computed above.
+       behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
+       deal with possessive ONCEs specially.
-       Then, when we are doing the actual compile phase, check to see whether
-       this group is a non-atomic one that could match an empty string. If so,
+       Otherwise, when we are doing the actual compile phase, check to see
+       whether this group is one that could match an empty string. If so,
        convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
-       that runtime checking can be done. [This check is also applied to
+       that runtime checking can be done. [This check is also applied to ONCE
-       atomic groups at runtime, but in a different way.] */
+       groups at runtime, but in a different way.]
+       Then, if the quantifier was possessive and the bracket is not a
+       conditional, we convert the BRA code to the POS form, and the KET code to
+       KETRPOS. (It turns out to be convenient at runtime to detect this kind of
+       subpattern at both the start and at the end.) The use of special opcodes
+       makes it possible to reduce greatly the stack usage in pcre_exec(). If
+       the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
+       Then, if the minimum number of matches is 1 or 0, cancel the possessive
+       flag so that the default action below, of wrapping everything inside
+       atomic brackets, does not happen. When the minimum is greater than 1,
+       there will be earlier copies of the group, and so we still have to wrap
+       the whole thing. */
        else
          {
-         uschar *ketcode = code - ketoffset;
+         uschar *ketcode = code - 1 - LINK_SIZE;
          uschar *bracode = ketcode - GET(ketcode, 1);
-         *ketcode = OP_KETRMAX + repeat_type;
-         if (lengthptr == NULL && *bracode != OP_ONCE)
+         /* Convert possessive ONCE brackets to non-capturing */
+         if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
+             possessive_quantifier) *bracode = OP_BRA;
+         /* For non-possessive ONCE brackets, all we need to do is to
+         set the KET. */
+         if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
+           *ketcode = OP_KETRMAX + repeat_type;
+         /* Handle non-ONCE brackets and possessive ONCEs (which have been
+         converted to non-capturing above). */
+         else
            {
-           uschar *scode = bracode;
+           /* In the compile phase, check for empty string matching. */
-           do
+           if (lengthptr == NULL)
              {
-             if (could_be_empty_branch(scode, ketcode, utf8, cd))
+             uschar *scode = bracode;
+             do
                {
-               *bracode += OP_SBRA - OP_BRA;
+               if (could_be_empty_branch(scode, ketcode, utf8, cd))
-               break;
+                 {
+                 *bracode += OP_SBRA - OP_BRA;
+                 break;
+                 }
+               scode += GET(scode, 1);
                }
-             scode += GET(scode, 1);
+             while (*scode == OP_ALT);
              }
-           while (*scode == OP_ALT);
+           /* Handle possessive quantifiers. */
+           if (possessive_quantifier)
+             {
+             /* For COND brackets, we wrap the whole thing in a possessively
+             repeated non-capturing bracket, because we have not invented POS
+             versions of the COND opcodes. Because we are moving code along, we
+             must ensure that any pending recursive references are updated. */
+             if (*bracode == OP_COND || *bracode == OP_SCOND)
+               {
+               int nlen = (int)(code - bracode);
+               *code = OP_END;
+               adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
+               memmove(bracode + 1+LINK_SIZE, bracode, nlen);
+               code += 1 + LINK_SIZE;
+               nlen += 1 + LINK_SIZE;
+               *bracode = OP_BRAPOS;
+               *code++ = OP_KETRPOS;
+               PUTINC(code, 0, nlen);
+               PUT(bracode, 1, nlen);
+               }
+             /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
+             else
+               {
+               *bracode += 1;              /* Switch to xxxPOS opcodes */
+               *ketcode = OP_KETRPOS;
+               }
+             /* If the minimum is zero, mark it as possessive, then unset the
+             possessive flag when the minimum is 0 or 1. */
+             if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
+             if (repeat_min < 2) possessive_quantifier = FALSE;
+             }
+           /* Non-possessive quantifier */
+           else *ketcode = OP_KETRMAX + repeat_type;
            }
          }
        }
-Line 4674 
 for (;; ptr++)
+Line 5188 
 for (;; ptr++)
        }
      /* If the character following a repeat is '+', or if certain optimization
-     tests above succeeded, possessive_quantifier is TRUE. For some of the
+     tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
-     simpler opcodes, there is an special alternative opcode for this. For
+     there are special alternative opcodes for this case. For anything else, we
-     anything else, we wrap the entire repeated item inside OP_ONCE brackets.
+     wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
-     The '+' notation is just syntactic sugar, taken from Sun's Java package,
+     notation is just syntactic sugar, taken from Sun's Java package, but the
-     but the special opcodes can optimize it a bit. The repeated item starts at
+     special opcodes can optimize it.
-     tempcode, not at previous, which might be the first part of a string whose
-     (former) last char we repeated.
+     Some (but not all) possessively repeated subpatterns have already been
+     completely handled in the code just above. For them, possessive_quantifier
+     is always FALSE at this stage.
+     Note that the repeated item starts at tempcode, not at previous, which
+     might be the first part of a string whose (former) last char we repeated.
      Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
      an 'upto' may follow. We skip over an 'exact' item, and then test the
-Line 4711 
 for (;; ptr++)
+Line 5230 
 for (;; ptr++)
          case OP_QUERY: *tempcode = OP_POSQUERY; break;
          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
-         case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
+         case OP_STARI:  *tempcode = OP_POSSTARI; break;
-         case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
+         case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
-         case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
+         case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
-         case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
+         case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
+         case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
+         case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
+         case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
+         case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
+         case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
+         case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
+         case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
+         case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
          /* Because we are moving code along, we must ensure that any
          pending recursive references are updated. */
-Line 4774 
 for (;; ptr++)
+Line 5303 
 for (;; ptr++)
        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
        namelen = (int)(ptr - name);
+       /* It appears that Perl allows any characters whatsoever, other than
+       a closing parenthesis, to appear in arguments, so we no longer insist on
+       letters, digits, and underscores. */
        if (*ptr == CHAR_COLON)
          {
          arg = ++ptr;
-         while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
+         while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
-           || *ptr == '_') ptr++;
          arglen = (int)(ptr - arg);
          }
-Line 4795 
 for (;; ptr++)
+Line 5327 
 for (;; ptr++)
          if (namelen == verbs[i].len &&
              strncmp((char *)name, vn, namelen) == 0)
            {
-           /* Check for open captures before ACCEPT */
+           /* Check for open captures before ACCEPT and convert it to
+           ASSERT_ACCEPT if in an assertion. */
            if (verbs[i].op == OP_ACCEPT)
              {
              open_capitem *oc;
+             if (arglen != 0)
+               {
+               *errorcodeptr = ERR59;
+               goto FAILED;
+               }
              cd->had_accept = TRUE;
              for (oc = cd->open_caps; oc != NULL; oc = oc->next)
                {
                *code++ = OP_CLOSE;
                PUT2INC(code, 0, oc->number);
                }
+             *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
+             /* Do not set firstbyte after *ACCEPT */
+             if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
              }
-           /* Handle the cases with/without an argument */
+           /* Handle other cases with/without an argument */
-           if (arglen == 0)
+           else if (arglen == 0)
              {
              if (verbs[i].op < 0)   /* Argument is mandatory */
                {
-Line 4818 
 for (;; ptr++)
+Line 5360 
 for (;; ptr++)
                goto FAILED;
                }
              *code = verbs[i].op;
-             if (*code++ == OP_THEN)
+             if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
-               {
-               PUT(code, 0, code - bcptr->current_branch - 1);
-               code += LINK_SIZE;
-               }
              }
            else
-Line 4833 
 for (;; ptr++)
+Line 5371 
 for (;; ptr++)
                goto FAILED;
                }
              *code = verbs[i].op_arg;
-             if (*code++ == OP_THEN_ARG)
+             if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN;
-               {
-               PUT(code, 0, code - bcptr->current_branch - 1);
-               code += LINK_SIZE;
-               }
              *code++ = arglen;
              memcpy(code, arg, arglen);
              code += arglen;
-Line 5031 
 for (;; ptr++)
+Line 5565 
 for (;; ptr++)
          /* Search the pattern for a forward reference */
          else if ((i = find_parens(cd, name, namelen,
-                         (options & PCRE_EXTENDED) != 0)) > 0)
+                         (options & PCRE_EXTENDED) != 0, utf8)) > 0)
            {
            PUT2(code, 2+LINK_SIZE, i);
            code[1+LINK_SIZE]++;
-Line 5099 
 for (;; ptr++)
+Line 5633 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
          bravalue = OP_ASSERT;
+         cd->assert_depth += 1;
          ptr++;
          break;
-Line 5113 
 for (;; ptr++)
+Line 5648 
 for (;; ptr++)
            continue;
            }
          bravalue = OP_ASSERT_NOT;
+         cd->assert_depth += 1;
          break;
-Line 5122 
 for (;; ptr++)
+Line 5658 
 for (;; ptr++)
            {
            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
            bravalue = OP_ASSERTBACK;
+           cd->assert_depth += 1;
            ptr += 2;
            break;
            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
            bravalue = OP_ASSERTBACK_NOT;
+           cd->assert_depth += 1;
            ptr += 2;
            break;
-Line 5148 
 for (;; ptr++)
+Line 5686 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
          case CHAR_C:                 /* Callout - may be followed by digits; */
-         previous_callout = code;  /* Save for later completion */
+         previous_callout = code;     /* Save for later completion */
-         after_manual_callout = 1; /* Skip one item before completing */
+         after_manual_callout = 1;    /* Skip one item before completing */
          *code++ = OP_CALLOUT;
            {
            int n = 0;
-Line 5332 
 for (;; ptr++)
+Line 5870 
 for (;; ptr++)
          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
          namelen = (int)(ptr - name);
-         /* In the pre-compile phase, do a syntax check and set a dummy
+         /* In the pre-compile phase, do a syntax check. We used to just set
-         reference number. */
+         a dummy reference number, because it was not used in the first pass.
+         However, with the change of recursive back references to be atomic,
+         we have to look for the number so that this state can be identified, as
+         otherwise the incorrect length is computed. If it's not a backwards
+         reference, the dummy number will do. */
          if (lengthptr != NULL)
            {
+           const uschar *temp;
            if (namelen == 0)
              {
              *errorcodeptr = ERR62;
-Line 5352 
 for (;; ptr++)
+Line 5896 
 for (;; ptr++)
              *errorcodeptr = ERR48;
              goto FAILED;
              }
-           recno = 0;
+           /* The name table does not exist in the first pass, so we cannot
+           do a simple search as in the code below. Instead, we have to scan the
+           pattern to find the number. It is important that we scan it only as
+           far as we have got because the syntax of named subpatterns has not
+           been checked for the rest of the pattern, and find_parens() assumes
+           correct syntax. In any case, it's a waste of resources to scan
+           further. We stop the scan at the current point by temporarily
+           adjusting the value of cd->endpattern. */
+           temp = cd->end_pattern;
+           cd->end_pattern = ptr;
+           recno = find_parens(cd, name, namelen,
+             (options & PCRE_EXTENDED) != 0, utf8);
+           cd->end_pattern = temp;
+           if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
            }
          /* In the real compile, seek the name in the table. We check the name
-Line 5377 
 for (;; ptr++)
+Line 5936 
 for (;; ptr++)
              }
            else if ((recno =                /* Forward back reference */
                      find_parens(cd, name, namelen,
-                       (options & PCRE_EXTENDED) != 0)) <= 0)
+                       (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
              {
              *errorcodeptr = ERR15;
              goto FAILED;
-Line 5488 
 for (;; ptr++)
+Line 6047 
 for (;; ptr++)
              if (called == NULL)
                {
                if (find_parens(cd, NULL, recno,
-                     (options & PCRE_EXTENDED) != 0) < 0)
+                     (options & PCRE_EXTENDED) != 0, utf8) < 0)
                  {
                  *errorcodeptr = ERR15;
                  goto FAILED;
-Line 5496 
 for (;; ptr++)
+Line 6055 
 for (;; ptr++)
                /* Fudge the value of "called" so that when it is inserted as an
                offset below, what it actually inserted is the reference number
-               of the group. */
+               of the group. Then remember the forward reference. */
                called = cd->start_code + recno;
-               PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
+               if (cd->hwm >= cd->start_workspace + cd->workspace_size -
+                   WORK_SIZE_SAFETY_MARGIN)
+                 {
+                 *errorcodeptr = expand_workspace(cd);
+                 if (*errorcodeptr != 0) goto FAILED;
+                 }
+               PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
                }
              /* If not a forward reference, and the subpattern is still open,
              this is a recursive call. We check to see if this is a left
-             recursion that could loop for ever, and diagnose that case. */
+             recursion that could loop for ever, and diagnose that case. We
+             must not, however, do this check if we are in a conditional
+             subpattern because the condition might be testing for recursion in
+             a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
+             Forever loops are also detected at runtime, so those that occur in
+             conditional subpatterns will be picked up then. */
-             else if (GET(called, 1) == 0 &&
+             else if (GET(called, 1) == 0 && cond_depth <= 0 &&
                       could_be_empty(called, code, bcptr, utf8, cd))
                {
                *errorcodeptr = ERR40;
-Line 5514 
 for (;; ptr++)
+Line 6084 
 for (;; ptr++)
                }
              }
-           /* Insert the recursion/subroutine item, automatically wrapped inside
+           /* Insert the recursion/subroutine item. It does not have a set first
-           "once" brackets. Set up a "previous group" length so that a
+           byte (relevant if it is repeated, because it will then be wrapped
-           subsequent quantifier will work. */
+           with ONCE brackets). */
-           *code = OP_ONCE;
-           PUT(code, 1, 2 + 2*LINK_SIZE);
-           code += 1 + LINK_SIZE;
            *code = OP_RECURSE;
            PUT(code, 1, (int)(called - cd->start_code));
            code += 1 + LINK_SIZE;
+           groupsetfirstbyte = FALSE;
-           *code = OP_KET;
-           PUT(code, 1, 2 + 2*LINK_SIZE);
-           code += 1 + LINK_SIZE;
-           length_prevgroup = 3 + 3*LINK_SIZE;
            }
          /* Can't determine a first byte now */
-Line 5591 
 for (;; ptr++)
+Line 6152 
 for (;; ptr++)
          is necessary to ensure we correctly detect the start of the pattern in
          both phases.
-         If we are not at the pattern start, compile code to change the ims
+         If we are not at the pattern start, reset the greedy defaults and the
-         options if this setting actually changes any of them, and reset the
+         case value for firstbyte and reqbyte. */
-         greedy defaults and the case value for firstbyte and reqbyte. */
          if (*ptr == CHAR_RIGHT_PARENTHESIS)
            {
-Line 5604 
 for (;; ptr++)
+Line 6164 
 for (;; ptr++)
              }
            else
              {
-             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
-               {
-               *code++ = OP_OPT;
-               *code++ = newoptions & PCRE_IMS;
-               }
              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
              greedy_non_default = greedy_default ^ 1;
              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
              }
            /* Change options at this level, and pass them back for use
-           in subsequent branches. When not at the start of the pattern, this
+           in subsequent branches. */
-           information is also necessary so that a resetting item can be
-           compiled at the end of a group (if we are in a group). */
            *optionsptr = options = newoptions;
            previous = NULL;       /* This item can't be repeated */
-Line 5653 
 for (;; ptr++)
+Line 6206 
 for (;; ptr++)
        skipbytes = 2;
        }
-     /* Process nested bracketed regex. Assertions may not be repeated, but
+     /* Process nested bracketed regex. Assertions used not to be repeatable,
-     other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
+     but this was changed for Perl compatibility, so all kinds can now be
-     non-register variable in order to be able to pass its address because some
+     repeated. We copy code into a non-register variable (tempcode) in order to
-     compilers complain otherwise. Pass in a new setting for the ims options if
+     be able to pass its address because some compilers complain otherwise. */
-     they have changed. */
-     previous = (bravalue >= OP_ONCE)? code : NULL;
+     previous = code;                      /* For handling repetition */
      *code = bravalue;
      tempcode = code;
-     tempreqvary = cd->req_varyopt;     /* Save value before bracket */
+     tempreqvary = cd->req_varyopt;        /* Save value before bracket */
-     length_prevgroup = 0;              /* Initialize for pre-compile phase */
+     tempbracount = cd->bracount;          /* Save value before bracket */
+     length_prevgroup = 0;                 /* Initialize for pre-compile phase */
      if (!compile_regex(
-          newoptions,                   /* The complete new option state */
+          newoptions,                      /* The complete new option state */
-          options & PCRE_IMS,           /* The previous ims option state */
+          &tempcode,                       /* Where to put code (updated) */
-          &tempcode,                    /* Where to put code (updated) */
+          &ptr,                            /* Input pointer (updated) */
-          &ptr,                         /* Input pointer (updated) */
+          errorcodeptr,                    /* Where to put an error message */
-          errorcodeptr,                 /* Where to put an error message */
           (bravalue == OP_ASSERTBACK ||
            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
-          reset_bracount,               /* True if (?| group */
+          reset_bracount,                  /* True if (?| group */
-          skipbytes,                    /* Skip over bracket number */
+          skipbytes,                       /* Skip over bracket number */
-          &subfirstbyte,                /* For possible first char */
+          cond_depth +
-          &subreqbyte,                  /* For possible last char */
+            ((bravalue == OP_COND)?1:0),   /* Depth of condition subpatterns */
-          bcptr,                        /* Current branch chain */
+          &subfirstbyte,                   /* For possible first char */
-          cd,                           /* Tables block */
+          &subreqbyte,                     /* For possible last char */
-          (lengthptr == NULL)? NULL :   /* Actual compile phase */
+          bcptr,                           /* Current branch chain */
-            &length_prevgroup           /* Pre-compile phase */
+          cd,                              /* Tables block */
+          (lengthptr == NULL)? NULL :      /* Actual compile phase */
+            &length_prevgroup              /* Pre-compile phase */
           ))
        goto FAILED;
+     /* If this was an atomic group and there are no capturing groups within it,
+     generate OP_ONCE_NC instead of OP_ONCE. */
+     if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
+       *code = OP_ONCE_NC;
+     if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
+       cd->assert_depth -= 1;
      /* At the end of compiling, code is still pointing to the start of the
-     group, while tempcode has been updated to point past the end of the group
+     group, while tempcode has been updated to point past the end of the group.
-     and any option resetting that may follow it. The pattern pointer (ptr)
+     The pattern pointer (ptr) is on the bracket.
-     is on the bracket. */
-     /* If this is a conditional bracket, check that there are no more than
+     If this is a conditional bracket, check that there are no more than
      two branches in the group, or just one if it's a DEFINE group. We do this
      in the real compile phase, not in the pre-pass, where the whole group may
      not be available. */
-Line 5754 
 for (;; ptr++)
+Line 6316 
 for (;; ptr++)
          goto FAILED;
          }
        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
-       *code++ = OP_BRA;
+       code++;   /* This already contains bravalue */
        PUTINC(code, 0, 1 + LINK_SIZE);
        *code++ = OP_KET;
        PUTINC(code, 0, 1 + LINK_SIZE);
-Line 5922 
 for (;; ptr++)
+Line 6484 
 for (;; ptr++)
          }
        /* \k<name> or \k'name' is a back reference by name (Perl syntax).
-       We also support \k{name} (.NET syntax) */
+       We also support \k{name} (.NET syntax).  */
-       if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
+       if (-c == ESC_k)
-           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
          {
+         if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
+           ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
+           {
+           *errorcodeptr = ERR69;
+           break;
+           }
          is_recurse = FALSE;
          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
-Line 5946 
 for (;; ptr++)
+Line 6513 
 for (;; ptr++)
          HANDLE_REFERENCE:    /* Come here from named backref handling */
          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
          previous = code;
-         *code++ = OP_REF;
+         *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
          PUT2INC(code, 0, recno);
          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
          if (recno > cd->top_backref) cd->top_backref = recno;
-Line 6006 
 for (;; ptr++)
+Line 6573 
 for (;; ptr++)
            }
          else
  #endif
-           {
+         /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
+         so that it works in DFA mode and in lookbehinds. */
+           {
            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
-           *code++ = -c;
+           *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
            }
          }
        continue;
-Line 6054 
 for (;; ptr++)
+Line 6624 
 for (;; ptr++)
      ONE_CHAR:
      previous = code;
-     *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
+     *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
      /* Remember if \r or \n were seen */
-Line 6083 
 for (;; ptr++)
+Line 6653 
 for (;; ptr++)
        else firstbyte = reqbyte = REQ_NONE;
        }
-     /* firstbyte was previously set; we can set reqbyte only the length is
+     /* firstbyte was previously set; we can set reqbyte only if the length is
 or the matching is caseful. */
      else
-Line 6118 
 return FALSE;
+Line 6688 
 return FALSE;
  /* On entry, ptr is pointing past the bracket character, but on return it
  points to the closing bracket, or vertical bar, or end of string. The code
  variable is pointing at the byte into which the BRA operator has been stored.
- If the ims options are changed at the start (for a (?ims: group) or during any
- branch, we need to insert an OP_OPT item at the start of every following branch
- to ensure they get set correctly at run time, and also pass the new options
- into every subsequent branch compile.
  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:
    options        option bits, including any changes for this subpattern
-   oldims         previous settings of ims option bits
    codeptr        -> the address of the current code pointer
    ptrptr         -> the address of the current pattern pointer
    errorcodeptr   -> pointer to error code variable
    lookbehind     TRUE if this is a lookbehind assertion
    reset_bracount TRUE to reset the count for each branch
    skipbytes      skip this many bytes at start (for brackets and OP_COND)
+   cond_depth     depth of nesting for conditional subpatterns
    firstbyteptr   place to put the first required character, or a negative number
    reqbyteptr     place to put the last required character, or a negative number
    bcptr          pointer to the chain of currently open branches
-Line 6147 
 Returns:         TRUE on success
+Line 6712 
 Returns:         TRUE on success
  */
  static BOOL
- compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
+ compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
-   int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
+   int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
-   int *lengthptr)
+   compile_data *cd, int *lengthptr)
  {
  const uschar *ptr = *ptrptr;
  uschar *code = *codeptr;
-Line 6164 
 int branchfirstbyte, branchreqbyte;
+Line 6729 
 int branchfirstbyte, branchreqbyte;
  int length;
  int orig_bracount;
  int max_bracount;
- int old_external_options = cd->external_options;
  branch_chain bc;
  bc.outer = bcptr;
-Line 6188 
 pre-compile phase to find out whether an
+Line 6752 
 pre-compile phase to find out whether an
  /* If this is a capturing subpattern, add to the chain of open capturing items
  so that we can detect them if (*ACCEPT) is encountered. This is also used to
- detect groups that contain recursive back references to themselves. */
+ detect groups that contain recursive back references to themselves. Note that
+ only OP_CBRA need be tested here; changing this opcode to one of its variants,
+ e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
  if (*code == OP_CBRA)
    {
-Line 6214 
 for (;;)
+Line 6780 
 for (;;)
    if (reset_bracount) cd->bracount = orig_bracount;
-   /* Handle a change of ims options at the start of the branch */
-   if ((options & PCRE_IMS) != oldims)
-     {
-     *code++ = OP_OPT;
-     *code++ = options & PCRE_IMS;
-     length += 2;
-     }
    /* Set up dummy OP_REVERSE if lookbehind assertion */
    if (lookbehind)
-Line 6237 
 for (;;)
+Line 6794 
 for (;;)
    into the length. */
    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
-         &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
+         &branchreqbyte, &bc, cond_depth, cd,
+         (lengthptr == NULL)? NULL : &length))
      {
      *ptrptr = ptr;
      return FALSE;
      }
-   /* If the external options have changed during this branch, it means that we
-   are at the top level, and a leading option setting has been encountered. We
-   need to re-set the original option values to take account of this so that,
-   during the pre-compile phase, we know to allow for a re-set at the start of
-   subsequent branches. */
-   if (old_external_options != cd->external_options)
-     oldims = cd->external_options & PCRE_IMS;
    /* Keep the highest bracket count in case (?| was used and some branch
    has fewer than the rest. */
-Line 6312 
 for (;;)
+Line 6861 
 for (;;)
        {
        int fixed_length;
        *code = OP_END;
-       fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
+       fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
+         FALSE, cd);
        DPRINTF(("fixed length = %d\n", fixed_length));
        if (fixed_length == -3)
          {
-Line 6320 
 for (;;)
+Line 6870 
 for (;;)
          }
        else if (fixed_length < 0)
          {
-         *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
+         *errorcodeptr = (fixed_length == -2)? ERR36 :
+                         (fixed_length == -4)? ERR70: ERR25;
          *ptrptr = ptr;
          return FALSE;
          }
-Line 6333 
 for (;;)
+Line 6884 
 for (;;)
    of offsets, with the field in the BRA item now becoming an offset to the
    first alternative. If there are no alternatives, it points to the end of the
    group. The length in the terminating ket is always the length of the whole
-   bracketed item. If any of the ims options were changed inside the group,
+   bracketed item. Return leaving the pointer at the terminating char. */
-   compile a resetting op-code following, except at the very end of the pattern.
-   Return leaving the pointer at the terminating char. */
    if (*ptr != CHAR_VERTICAL_LINE)
      {
-Line 6379 
 for (;;)
+Line 6928 
 for (;;)
        cd->open_caps = cd->open_caps->next;
        }
-     /* Reset options if needed. */
-     if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
-       {
-       *code++ = OP_OPT;
-       *code++ = oldims;
-       length += 2;
-       }
      /* Retain the highest bracket number, in case resetting was used. */
      cd->bracount = max_bracount;
-Line 6447 
 for (;;)
+Line 6987 
 for (;;)
  /* Try to find out if this is an anchored regular expression. Consider each
  alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
  all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
- it's anchored. However, if this is a multiline pattern, then only OP_SOD
+ it's anchored. However, if this is a multiline pattern, then only OP_SOD will
- counts, since OP_CIRC can match in the middle.
+ be found, because ^ generates OP_CIRCM in that mode.
  We can also consider a regex to be anchored if OP_SOM starts all its branches.
  This is the code for \G, which means "match at start of match position, taking
-Line 6469 
 of the more common cases more precisely.
+Line 7009 
 of the more common cases more precisely.
  Arguments:
    code           points to start of expression (the bracket)
-   options        points to the options setting
    bracket_map    a bitmap of which brackets we are inside while testing; this
                    handles up to substring 31; after that we just have to take
                    the less precise approach
-Line 6479 
 Returns:     TRUE or FALSE
+Line 7018 
 Returns:     TRUE or FALSE
  */
  static BOOL
- is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
+ is_anchored(register const uschar *code, unsigned int bracket_map,
    unsigned int backref_map)
  {
  do {
     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
-      options, PCRE_MULTILINE, FALSE);
+      FALSE);
     register int op = *scode;
     /* Non-capturing brackets */
-    if (op == OP_BRA)
+    if (op == OP_BRA  || op == OP_BRAPOS ||
+        op == OP_SBRA || op == OP_SBRAPOS)
       {
-      if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
+      if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
       }
     /* Capturing brackets */
-    else if (op == OP_CBRA)
+    else if (op == OP_CBRA  || op == OP_CBRAPOS ||
+             op == OP_SCBRA || op == OP_SCBRAPOS)
       {
       int n = GET2(scode, 1+LINK_SIZE);
       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
-      if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
+      if (!is_anchored(scode, new_map, backref_map)) return FALSE;
       }
     /* Other brackets */
-    else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
+    else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC ||
+             op == OP_COND)
       {
-      if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
+      if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
       }
     /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
-Line 6522 
 do {
+Line 7064 
 do {
     /* Check for explicit anchoring */
-    else if (op != OP_SOD && op != OP_SOM &&
+    else if (op != OP_SOD && op != OP_SOM && op != OP_CIRC) return FALSE;
-            ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
-      return FALSE;
     code += GET(code, 1);
     }
  while (*code == OP_ALT);   /* Loop for each alternative */
-Line 6560 
 is_startline(const uschar *code, unsigne
+Line 7100 
 is_startline(const uschar *code, unsigne
  {
  do {
     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
-      NULL, 0, FALSE);
+      FALSE);
     register int op = *scode;
     /* If we are at the start of a conditional assertion group, *both* the
-Line 6587 
 do {
+Line 7127 
 do {
         scode += 1 + LINK_SIZE;
         break;
         }
-      scode = first_significant_code(scode, NULL, 0, FALSE);
+      scode = first_significant_code(scode, FALSE);
       op = *scode;
       }
     /* Non-capturing brackets */
-    if (op == OP_BRA)
+    if (op == OP_BRA  || op == OP_BRAPOS ||
+        op == OP_SBRA || op == OP_SBRAPOS)
       {
       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
       }
     /* Capturing brackets */
-    else if (op == OP_CBRA)
+    else if (op == OP_CBRA  || op == OP_CBRAPOS ||
+             op == OP_SCBRA || op == OP_SCBRAPOS)
       {
       int n = GET2(scode, 1+LINK_SIZE);
       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
-Line 6609 
 do {
+Line 7151 
 do {
     /* Other brackets */
-    else if (op == OP_ASSERT || op == OP_ONCE)
+    else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC)
       {
       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
       }
-Line 6624 
 do {
+Line 7166 
 do {
     /* Check for explicit circumflex */
-    else if (op != OP_CIRC) return FALSE;
+    else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
     /* Move on to the next alternative */
-Line 6650 
 we return that char, otherwise -1.
+Line 7192 
 we return that char, otherwise -1.
  Arguments:
    code       points to start of expression (the bracket)
-   options    pointer to the options (used to check casing changes)
    inassert   TRUE if in an assertion
  Returns:     -1 or the fixed first char
  */
  static int
- find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
+ find_firstassertedchar(const uschar *code, BOOL inassert)
  {
  register int c = -1;
  do {
     int d;
-    const uschar *scode =
+    int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
-      first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
+              *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? 2:0;
+    const uschar *scode = first_significant_code(code + 1+LINK_SIZE + xl, TRUE);
     register int op = *scode;
     switch(op)
-Line 6672 
 do {
+Line 7214 
 do {
       return -1;
       case OP_BRA:
+      case OP_BRAPOS:
       case OP_CBRA:
+      case OP_SCBRA:
+      case OP_CBRAPOS:
+      case OP_SCBRAPOS:
       case OP_ASSERT:
       case OP_ONCE:
+      case OP_ONCE_NC:
       case OP_COND:
-      if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
+      if ((d = find_firstassertedchar(scode, op == OP_ASSERT)) < 0)
         return -1;
       if (c < 0) c = d; else if (c != d) return -1;
       break;
-      case OP_EXACT:       /* Fall through */
+      case OP_EXACT:
       scode += 2;
+      /* Fall through */
       case OP_CHAR:
-      case OP_CHARNC:
       case OP_PLUS:
       case OP_MINPLUS:
       case OP_POSPLUS:
       if (!inassert) return -1;
-      if (c < 0)
+      if (c < 0) c = scode[1];
-        {
+        else if (c != scode[1]) return -1;
-        c = scode[1];
+      break;
-        if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
-        }
+      case OP_EXACTI:
-      else if (c != scode[1]) return -1;
+      scode += 2;
+      /* Fall through */
+      case OP_CHARI:
+      case OP_PLUSI:
+      case OP_MINPLUSI:
+      case OP_POSPLUSI:
+      if (!inassert) return -1;
+      if (c < 0) c = scode[1] | REQ_CASELESS;
+        else if (c != scode[1]) return -1;
       break;
       }
-Line 6758 
 compile_data *cd = &compile_block;
+Line 7314 
 compile_data *cd = &compile_block;
  computing the amount of memory that is needed. Compiled items are thrown away
  as soon as possible, so that a fairly large buffer should be sufficient for
  this purpose. The same space is used in the second phase for remembering where
- to fill in forward references to subpatterns. */
+ to fill in forward references to subpatterns. That may overflow, in which case
+ new memory is obtained from malloc(). */
  uschar cworkspace[COMPILE_WORK_SIZE];
-Line 6818 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
+Line 7375 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
      { skipatstart += 7; options |= PCRE_UTF8; continue; }
    else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
      { skipatstart += 6; options |= PCRE_UCP; continue; }
+   else if (strncmp((char *)(ptr+skipatstart+2), STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
+     { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
-Line 6844 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
+Line 7403 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
  utf8 = (options & PCRE_UTF8) != 0;
- /* Can't support UTF8 unless PCRE has been compiled to include the code. */
+ /* Can't support UTF8 unless PCRE has been compiled to include the code. The
+ return of an error code from _pcre_valid_utf8() is a new feature, introduced in
+ release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
+ not used here. */
  #ifdef SUPPORT_UTF8
  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
-      (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
+      (errorcode = _pcre_valid_utf8((USPTR)pattern, -1, erroroffset)) != 0)
    {
    errorcode = ERR44;
    goto PCRE_EARLY_ERROR_RETURN2;
-Line 6873 
 if ((options & PCRE_UCP) != 0)
+Line 7435 
 if ((options & PCRE_UCP) != 0)
  /* Check validity of \R options. */
- switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
+ if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
+      (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
    {
-   case 0:
+   errorcode = ERR56;
-   case PCRE_BSR_ANYCRLF:
+   goto PCRE_EARLY_ERROR_RETURN;
-   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
-Line 6945 
 cd->bracount = cd->final_bracount = 0;
+Line 7505 
 cd->bracount = cd->final_bracount = 0;
  cd->names_found = 0;
  cd->name_entry_size = 0;
  cd->name_table = NULL;
- cd->start_workspace = cworkspace;
  cd->start_code = cworkspace;
  cd->hwm = cworkspace;
+ cd->start_workspace = cworkspace;
+ cd->workspace_size = COMPILE_WORK_SIZE;
  cd->start_pattern = (const uschar *)pattern;
  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
  cd->req_varyopt = 0;
-Line 6964 
 outside can help speed up starting point
+Line 7525 
 outside can help speed up starting point
  ptr += skipatstart;
  code = cworkspace;
  *code = OP_BRA;
- (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
+ (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
-   &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
+   FALSE, 0, 0, &firstbyte, &reqbyte, NULL, cd, &length);
-   &length);
  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
-Line 7020 
 field; this time it's used for rememberi
+Line 7580 
 field; this time it's used for rememberi
  */
  cd->final_bracount = cd->bracount;  /* Save for checking forward references */
+ cd->assert_depth = 0;
  cd->bracount = 0;
  cd->names_found = 0;
  cd->name_table = (uschar *)re + re->name_table_offset;
  codestart = cd->name_table + re->name_entry_size * re->name_count;
  cd->start_code = codestart;
- cd->hwm = cworkspace;
+ cd->hwm = (uschar *)(cd->start_workspace);
  cd->req_varyopt = 0;
  cd->had_accept = FALSE;
  cd->check_lookbehind = FALSE;
-Line 7038 
 of the function here. */
+Line 7599 
 of the function here. */
  ptr = (const uschar *)pattern + skipatstart;
  code = (uschar *)codestart;
  *code = OP_BRA;
- (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
+ (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
-   &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
+   &firstbyte, &reqbyte, NULL, cd, NULL);
  re->top_bracket = cd->bracount;
  re->top_backref = cd->top_backref;
  re->flags = cd->external_flags;
- if (cd->had_accept) reqbyte = -1;   /* Must disable after (*ACCEPT) */
+ if (cd->had_accept) reqbyte = REQ_NONE;   /* Must disable after (*ACCEPT) */
  /* If not reached end of pattern on success, there's an excess bracket. */
-Line 7059 
 if debugging, leave the test till after
+Line 7620 
 if debugging, leave the test till after
  if (code - codestart > length) errorcode = ERR23;
  #endif
- /* Fill in any forward references that are required. */
+ /* Fill in any forward references that are required. There may be repeated
+ references; optimize for them, as searching a large regex takes time. */
- while (errorcode == 0 && cd->hwm > cworkspace)
+ if (cd->hwm > cd->start_workspace)
    {
-   int offset, recno;
+   int prev_recno = -1;
-   const uschar *groupptr;
+   const uschar *groupptr = NULL;
-   cd->hwm -= LINK_SIZE;
+   while (errorcode == 0 && cd->hwm > cd->start_workspace)
-   offset = GET(cd->hwm, 0);
+     {
-   recno = GET(codestart, offset);
+     int offset, recno;
-   groupptr = _pcre_find_bracket(codestart, utf8, recno);
+     cd->hwm -= LINK_SIZE;
-   if (groupptr == NULL) errorcode = ERR53;
+     offset = GET(cd->hwm, 0);
-     else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
+     recno = GET(codestart, offset);
-   }
+     if (recno != prev_recno)
+       {
+       groupptr = _pcre_find_bracket(codestart, utf8, recno);
+       prev_recno = recno;
+       }
+     if (groupptr == NULL) errorcode = ERR53;
+       else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
+     }
+   }
+ /* If the workspace had to be expanded, free the new memory. */
+ if (cd->workspace_size > COMPILE_WORK_SIZE)
+   (pcre_free)((void *)cd->start_workspace);
  /* Give an error if there's back reference to a non-existent capturing
  subpattern. */
-Line 7105 
 if (cd->check_lookbehind)
+Line 7680 
 if (cd->check_lookbehind)
        uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
        int end_op = *be;
        *be = OP_END;
-       fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
+       fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
+         cd);
        *be = end_op;
        DPRINTF(("fixed length = %d\n", fixed_length));
        if (fixed_length < 0)
          {
-         errorcode = (fixed_length == -2)? ERR36 : ERR25;
+         errorcode = (fixed_length == -2)? ERR36 :
+                     (fixed_length == -4)? ERR70 : ERR25;
          break;
          }
        PUT(cc, 1, fixed_length);
-Line 7144 
 start with ^. and also when all branches
+Line 7721 
 start with ^. and also when all branches
  if ((re->options & PCRE_ANCHORED) == 0)
    {
-   int temp_options = re->options;   /* May get changed during these scans */
+   if (is_anchored(codestart, 0, cd->backref_map))
-   if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
      re->options |= PCRE_ANCHORED;
    else
      {
      if (firstbyte < 0)
-       firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
+       firstbyte = find_firstassertedchar(codestart, FALSE);
      if (firstbyte >= 0)   /* Remove caseless flag for non-caseable chars */
        {
        int ch = firstbyte & 255;

 Legend:



Removed from v.550
 


changed lines


 
Added in v.773
 Legend:



Removed from v.550
 


changed lines


 
Added in v.773
-Removed from v.550
+Added in v.773

webmaster@exim.org	ViewVC Help
Powered by ViewVC 1.1.12