/[pcre]/code/branches/pcre16/pcre_compile.c

Diff of /code/branches/pcre16/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 530 by ph10,
Tue Jun  1 13:42:06 2010 UTC
+revision 618 by ph10,
Sat Jul 16 17:24:16 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 261 
 static const int posix_class_maps[] = {
    cbit_xdigit,-1,          0              /* xdigit */
  };
  /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
  substitutes must be in the order of the names, defined above, and there are
  both positive and negative cases. NULL means no substitute. */
  #ifdef SUPPORT_UCP
 Line 272 
 static const uschar *substitutes[] = {
    (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
    (uschar *)"\\p{Xsp}",   /* \s */
    (uschar *)"\\P{Xwd}",   /* \W */
    (uschar *)"\\p{Xwd}"    /* \w */
  };
  static const uschar *posix_substitutes[] = {
    (uschar *)"\\p{L}",     /* alpha */
    (uschar *)"\\p{Ll}",    /* lower */
    (uschar *)"\\p{Lu}",    /* upper */
    (uschar *)"\\p{Xan}",   /* alnum */
    NULL,                   /* ascii */
    (uschar *)"\\h",        /* blank */
    NULL,                   /* cntrl */
 Line 289 
 static const uschar *posix_substitutes[]
    NULL,                   /* punct */
    (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
    (uschar *)"\\p{Xwd}",   /* word */
    NULL,                   /* xdigit */
    /* Negated cases */
    (uschar *)"\\P{L}",     /* ^alpha */
    (uschar *)"\\P{Ll}",    /* ^lower */
    (uschar *)"\\P{Lu}",    /* ^upper */
    (uschar *)"\\P{Xan}",   /* ^alnum */
    NULL,                   /* ^ascii */
    (uschar *)"\\H",        /* ^blank */
    NULL,                   /* ^cntrl */
 Line 304 
 static const uschar *posix_substitutes[]
    NULL,                   /* ^punct */
    (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
    (uschar *)"\\P{Xwd}",   /* ^word */
    NULL                    /* ^xdigit */
  };
  #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
  #endif
  #define STRING(a)  # a
  #define XSTRING(s) STRING(s)
 Line 407 
 static const char error_texts[] =
    /* 65 */
    "different names for subpatterns of the same number are not allowed\0"
    "(*MARK) must have an argument\0"
    "this version of PCRE is not compiled with PCRE_UCP support\0"
+   "\\c must be followed by an ASCII character\0"
    ;
  /* Table to identify digits and hex digits. This is used when compiling
-Line 544 
 static const unsigned char ebcdic_charta
+Line 545 
 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 *);
+     int *, branch_chain *, compile_data *, int *);
-Line 841 
 else
+Line 842 
 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 853 
 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 1099 
 top-level call starts at the beginning o
+Line 1105 
 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 1127 
 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 1135 
 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 1129 
 dealing with. The very first call may no
+Line 1147 
 dealing with. The very first call may no
  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
    {
-   if (ptr[1] == CHAR_QUESTION_MARK &&
+   /* Handle specials such as (*SKIP) or (*UTF8) etc. */
-       ptr[2] == CHAR_VERTICAL_LINE)
+   if (ptr[1] == CHAR_ASTERISK) ptr += 2;
+   /* Handle a normal, unnamed capturing parenthesis. */
+   else if (ptr[1] != CHAR_QUESTION_MARK)
+     {
+     *count += 1;
+     if (name == NULL && *count == lorn) return *count;
+     ptr++;
+     }
+   /* All cases now have (? at the start. Remember when we are in a group
+   where the parenthesis numbers are duplicated. */
+   else if (ptr[2] == CHAR_VERTICAL_LINE)
      {
      ptr += 3;
      dup_parens = TRUE;
      }
-   /* Handle a normal, unnamed capturing parenthesis */
+   /* Handle comments; all characters are allowed until a ket is reached. */
-   else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
+   else if (ptr[2] == CHAR_NUMBER_SIGN)
      {
-     *count += 1;
+     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
-     if (name == NULL && *count == lorn) return *count;
+     goto FAIL_EXIT;
-     ptr++;
      }
    /* Handle a condition. If it is an assertion, just carry on so that it
    is processed as normal. If not, skip to the closing parenthesis of the
-   condition (there can't be any nested parens. */
+   condition (there can't be any nested parens). */
    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
      {
-Line 1159 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1191 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
        }
      }
-   /* We have either (? or (* and not a condition */
+   /* Start with (? but not a condition. */
    else
      {
-Line 1188 
 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
+Line 1220 
 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 1264 
 for (; *ptr != 0; ptr++)
+Line 1298 
 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 1273 
 for (; *ptr != 0; ptr++)
+Line 1315 
 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 1281 
 for (; *ptr != 0; ptr++)
+Line 1323 
 for (; *ptr != 0; ptr++)
    else if (*ptr == CHAR_RIGHT_PARENTHESIS)
      {
      if (dup_parens && *count < hwm_count) *count = hwm_count;
-     *ptrptr = ptr;
+     goto FAIL_EXIT;
-     return -1;
      }
    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
-Line 1320 
 Arguments:
+Line 1361 
 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 1338 
 matching closing parens. That is why we
+Line 1381 
 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 1354 
 return rc;
+Line 1397 
 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 1431 
 and doing the check at the end; a flag s
+Line 1464 
 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
-Line 1442 
 Returns:   the fixed length,
+Line 1475 
 Returns:   the fixed length,
  */
  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 1459 
 for (;;)
+Line 1492 
 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. They will cause a -1 response from the default
+     case of this switch. */
      case OP_CBRA:
      case OP_BRA:
      case OP_ONCE:
      case OP_COND:
-     d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), 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);
-Line 1472 
 for (;;)
+Line 1511 
 for (;;)
      /* Reached end of a branch; if it's a ket it is the end of a nested
      call. If it's ALT it is an alternation in a nested call. If it is
-     END it's the end of the outer call. All can be handled by the same code. */
+     END it's the end of the outer call. All can be handled by the same code.
+     Note that we must not include the OP_KETRxxx opcodes here, because they
+     all imply an unlimited repeat. */
      case OP_ALT:
      case OP_KET:
-     case OP_KETRMAX:
-     case OP_KETRMIN:
      case OP_END:
      if (length < 0) length = branchlength;
        else if (length != branchlength) return -1;
-Line 1495 
 for (;;)
+Line 1534 
 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 1518 
 for (;;)
+Line 1557 
 for (;;)
      case OP_RREF:
      case OP_NRREF:
      case OP_DEF:
-     case OP_OPT:
      case OP_CALLOUT:
      case OP_SOD:
      case OP_SOM:
-Line 1526 
 for (;;)
+Line 1564 
 for (;;)
      case OP_EOD:
      case OP_EODN:
      case OP_CIRC:
+     case OP_CIRCM:
      case OP_DOLL:
+     case OP_DOLLM:
      case OP_NOT_WORD_BOUNDARY:
      case OP_WORD_BOUNDARY:
      cc += _pcre_OP_lengths[*cc];
-Line 1535 
 for (;;)
+Line 1575 
 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 1552 
 for (;;)
+Line 1592 
 for (;;)
      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 1655 
 _pcre_find_bracket(const uschar *code, B
+Line 1694 
 _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 1673 
 for (;;)
+Line 1713 
 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 1711 
 for (;;)
+Line 1752 
 for (;;)
        case OP_MARK:
        case OP_PRUNE_ARG:
        case OP_SKIP_ARG:
-       case OP_THEN_ARG:
        code += code[1];
        break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       break;
        }
      /* Add in the fixed length from the table */
-Line 1728 
 for (;;)
+Line 1772 
 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 1814 
 for (;;)
+Line 1871 
 for (;;)
        case OP_MARK:
        case OP_PRUNE_ARG:
        case OP_SKIP_ARG:
-       case OP_THEN_ARG:
        code += code[1];
        break;
+       case OP_THEN_ARG:
+       code += code[1+LINK_SIZE];
+       break;
        }
      /* Add in the fixed length from the table */
-Line 1831 
 for (;;)
+Line 1891 
 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 1883 
 could_be_empty_branch(const uschar *code
+Line 1956 
 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 1901 
 for (code = first_significant_code(code
+Line 1974 
 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. */
-Line 1933 
 for (code = first_significant_code(code
+Line 1996 
 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 ||
+       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_COND)
      {
      BOOL empty_branch;
      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
-Line 2022 
 for (code = first_significant_code(code
+Line 2109 
 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 2063 
 for (code = first_significant_code(code
+Line 2151 
 for (code = first_significant_code(code
      case OP_KET:
      case OP_KETRMAX:
      case OP_KETRMIN:
+     case OP_KETRPOS:
      case OP_ALT:
      return TRUE;
-Line 2071 
 for (code = first_significant_code(code
+Line 2160 
 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 2092 
 for (code = first_significant_code(code
+Line 2190 
 for (code = first_significant_code(code
      case OP_MARK:
      case OP_PRUNE_ARG:
      case OP_SKIP_ARG:
-     case OP_THEN_ARG:
      code += code[1];
      break;
+     case OP_THEN_ARG:
+     code += code[1+LINK_SIZE];
+     break;
      /* None of the remaining opcodes are required to match a character. */
      default:
-Line 2392 
 for (++c; c <= d; c++)
+Line 2493 
 for (++c; c <= d; c++)
  return TRUE;
  }
+ /*************************************************
+ *        Check a character and a property        *
+ *************************************************/
+ /* This function is called by check_auto_possessive() when a property item
+ is adjacent to a fixed character.
+ Arguments:
+   c            the character
+   ptype        the property type
+   pdata        the data for the type
+   negated      TRUE if it's a negated property (\P or \p{^)
+ Returns:       TRUE if auto-possessifying is OK
+ */
+ static BOOL
+ check_char_prop(int c, int ptype, int pdata, BOOL negated)
+ {
+ const ucd_record *prop = GET_UCD(c);
+ switch(ptype)
+   {
+   case PT_LAMP:
+   return (prop->chartype == ucp_Lu ||
+           prop->chartype == ucp_Ll ||
+           prop->chartype == ucp_Lt) == negated;
+   case PT_GC:
+   return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
+   case PT_PC:
+   return (pdata == prop->chartype) == negated;
+   case PT_SC:
+   return (pdata == prop->script) == negated;
+   /* These are specials */
+   case PT_ALNUM:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
+   case PT_SPACE:    /* Perl space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_PXSPACE:  /* POSIX space */
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
+           c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
+           c == CHAR_FF || c == CHAR_CR)
+           == negated;
+   case PT_WORD:
+   return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
+           _pcre_ucp_gentype[prop->chartype] == ucp_N ||
+           c == CHAR_UNDERSCORE) == negated;
+   }
+ return FALSE;
+ }
  #endif  /* SUPPORT_UCP */
-Line 2405 
 whether the next thing could possibly ma
+Line 2569 
 whether the next thing could possibly ma
  sense to automatically possessify the repeated item.
  Arguments:
-   op_code       the repeated op code
+   previous      pointer to the repeated opcode
-   this          data for this item, depends on the opcode
    utf8          TRUE in UTF-8 mode
-   utf8_char     used for utf8 character bytes, NULL if not relevant
    ptr           next character in pattern
    options       options bits
    cd            contains pointers to tables etc.
-Line 2417 
 Returns:        TRUE if possessifying is
+Line 2579 
 Returns:        TRUE if possessifying is
  */
  static BOOL
- check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
+ check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
-   const uschar *ptr, int options, compile_data *cd)
+   int options, compile_data *cd)
  {
- int next;
+ int c, next;
+ int op_code = *previous++;
  /* Skip whitespace and comments in extended mode */
-Line 2431 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2594 
 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 2468 
 if ((options & PCRE_EXTENDED) != 0)
+Line 2638 
 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 2481 
 if (*ptr == CHAR_ASTERISK || *ptr == CHA
+Line 2658 
 if (*ptr == CHAR_ASTERISK || *ptr == CHA
    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
      return FALSE;
- /* Now compare the next item with the previous opcode. If the previous is a
+ /* Now compare the next item with the previous opcode. First, handle cases when
- positive single character match, "item" either contains the character or, if
+ the next item is a character. */
- "item" is greater than 127 in utf8 mode, the character's bytes are in
- utf8_char. */
- /* Handle cases when the next item is a character. */
  if (next >= 0) switch(op_code)
    {
    case OP_CHAR:
  #ifdef SUPPORT_UTF8
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
  #else
-   (void)(utf8_char);  /* Keep compiler happy by referencing function argument */
+   c = *previous;
  #endif
-   return item != next;
+   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
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
-   if (item == next) return FALSE;
+   if (c == next) return FALSE;
  #ifdef SUPPORT_UTF8
    if (utf8)
      {
-Line 2518 
 if (next >= 0) switch(op_code)
+Line 2692 
 if (next >= 0) switch(op_code)
  #else
      othercase = NOTACHAR;
  #endif
-     return (unsigned int)item != othercase;
+     return (unsigned int)c != othercase;
      }
    else
  #endif  /* SUPPORT_UTF8 */
-   return (item != cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
-   /* For OP_NOT, "item" must be 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:
-   if (item == next) return TRUE;
+   return (c = *previous) == next;
-   if ((options & PCRE_CASELESS) == 0) return FALSE;
+   case OP_NOTI:
+   if ((c = *previous) == next) return TRUE;
  #ifdef SUPPORT_UTF8
    if (utf8)
      {
-Line 2539 
 if (next >= 0) switch(op_code)
+Line 2717 
 if (next >= 0) switch(op_code)
  #else
      othercase = NOTACHAR;
  #endif
-     return (unsigned int)item == othercase;
+     return (unsigned int)c == othercase;
      }
    else
  #endif  /* SUPPORT_UTF8 */
-   return (item == cd->fcc[next]);  /* Non-UTF-8 mode */
+   return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
    /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
    When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
    case OP_DIGIT:
    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
-Line 2611 
 if (next >= 0) switch(op_code)
+Line 2789 
 if (next >= 0) switch(op_code)
      return op_code != OP_NOT_VSPACE;
      }
+ #ifdef SUPPORT_UCP
+   case OP_PROP:
+   return check_char_prop(next, previous[0], previous[1], FALSE);
+   case OP_NOTPROP:
+   return check_char_prop(next, previous[0], previous[1], TRUE);
+ #endif
    default:
    return FALSE;
    }
  /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
  is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
  generated only when PCRE_UCP is *not* set, that is, when only ASCII
- characteristics are recognized. */
+ characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
+ replaced by OP_PROP codes when PCRE_UCP is set. */
  switch(op_code)
    {
    case OP_CHAR:
-   case OP_CHARNC:
+   case OP_CHARI:
  #ifdef SUPPORT_UTF8
-   if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
+   GETCHARTEST(c, previous);
+ #else
+   c = *previous;
  #endif
    switch(-next)
      {
      case ESC_d:
-     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
      case ESC_D:
-     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
      case ESC_s:
-     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
      case ESC_S:
-     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
      case ESC_w:
-     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
+     return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
      case ESC_W:
-     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
+     return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
      case ESC_h:
      case ESC_H:
-     switch(item)
+     switch(c)
        {
        case 0x09:
        case 0x20:
-Line 2678 
 switch(op_code)
+Line 2867 
 switch(op_code)
      case ESC_v:
      case ESC_V:
-     switch(item)
+     switch(c)
        {
        case 0x0a:
        case 0x0b:
-Line 2692 
 switch(op_code)
+Line 2881 
 switch(op_code)
        return -next == ESC_v;
        }
+     /* When PCRE_UCP is set, these values get generated for \d etc. Find
+     their substitutions and process them. The result will always be either
+     -ESC_p or -ESC_P. Then fall through to process those values. */
+ #ifdef SUPPORT_UCP
+     case ESC_du:
+     case ESC_DU:
+     case ESC_wu:
+     case ESC_WU:
+     case ESC_su:
+     case ESC_SU:
+       {
+       int temperrorcode = 0;
+       ptr = substitutes[-next - ESC_DU];
+       next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
+       if (temperrorcode != 0) return FALSE;
+       ptr++;    /* For compatibility */
+       }
+     /* Fall through */
+     case ESC_p:
+     case ESC_P:
+       {
+       int ptype, pdata, errorcodeptr;
+       BOOL negated;
+       ptr--;      /* Make ptr point at the p or P */
+       ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
+       if (ptype < 0) return FALSE;
+       ptr++;      /* Point past the final curly ket */
+       /* If the property item is optional, we have to give up. (When generated
+       from \d etc by PCRE_UCP, this test will have been applied much earlier,
+       to the original \d etc. At this point, ptr will point to a zero byte. */
+       if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
+         strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
+           return FALSE;
+       /* Do the property check. */
+       return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
+       }
+ #endif
      default:
      return FALSE;
      }
+   /* In principle, support for Unicode properties should be integrated here as
+   well. It means re-organizing the above code so as to get hold of the property
+   values before switching on the op-code. However, I wonder how many patterns
+   combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
+   these op-codes are never generated.) */
    case OP_DIGIT:
    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
           next == -ESC_h || next == -ESC_v || next == -ESC_R;
-Line 2710 
 switch(op_code)
+Line 2950 
 switch(op_code)
    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
    case OP_HSPACE:
    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
           next == -ESC_w || next == -ESC_v || next == -ESC_R;
    case OP_NOT_HSPACE:
    return next == -ESC_h;
    /* Can't have \S in here because VT matches \S (Perl anomaly) */
    case OP_ANYNL:
    case OP_VSPACE:
    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
-Line 2725 
 switch(op_code)
+Line 2965 
 switch(op_code)
    return next == -ESC_v || next == -ESC_R;
    case OP_WORDCHAR:
    return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
           next == -ESC_v || next == -ESC_R;
    case OP_NOT_WORDCHAR:
-Line 2861 
 for (;; ptr++)
+Line 3101 
 for (;; ptr++)
    c = *ptr;
    /* If we are at the end of a nested substitution, revert to the outer level
    string. Nesting only happens one level deep. */
    if (c == 0 && nestptr != NULL)
-Line 2983 
 for (;; ptr++)
+Line 3223 
 for (;; ptr++)
      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 3030 
 for (;; ptr++)
+Line 3275 
 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 3168 
 for (;; ptr++)
+Line 3414 
 for (;; ptr++)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
        /* In the pre-compile phase, accumulate the length of any UTF-8 extra
        data and reset the pointer. This is so that very large classes that
        contain a zillion UTF-8 characters no longer overwrite the work space
-Line 3237 
 for (;; ptr++)
+Line 3483 
 for (;; ptr++)
          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
            posix_class = 0;
          /* When PCRE_UCP is set, some of the POSIX classes are converted to
          different escape sequences that use Unicode properties. */
  #ifdef SUPPORT_UCP
          if ((options & PCRE_UCP) != 0)
            {
            int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
            if (posix_substitutes[pc] != NULL)
              {
              nestptr = tempptr + 1;
              ptr = posix_substitutes[pc] - 1;
              continue;
              }
            }
  #endif
          /* In the non-UCP case, we build the bit map for the POSIX class in a
          chunk of local store because we may be adding and subtracting from it,
          and we don't want to subtract bits that may be in the main map already.
-Line 3339 
 for (;; ptr++)
+Line 3585 
 for (;; ptr++)
              case ESC_SU:
              nestptr = ptr;
              ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
              class_charcount -= 2;                /* Undo! */
              continue;
  #endif
              case ESC_d:
-Line 3360 
 for (;; ptr++)
+Line 3606 
 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 3781 
 for (;; ptr++)
+Line 4032 
 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 &&
        (!utf8 || !negate_class || class_lastchar < 128))
-Line 3800 
 for (;; ptr++)
+Line 4051 
 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 3870 
 for (;; ptr++)
+Line 4121 
 for (;; ptr++)
        }
  #endif
      /* If there are no characters > 255, or they are all to be included or
      excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
      whole class was negated and whether there were negative specials such as \S
      (non-UCP) in the class. Then copy the 32-byte map into the code vector,
-Line 3964 
 for (;; ptr++)
+Line 4215 
 for (;; ptr++)
      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 3998 
 for (;; ptr++)
+Line 4251 
 for (;; ptr++)
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
-             options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 4011 
 for (;; ptr++)
+Line 4263 
 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 &&
-           check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 4044 
 for (;; ptr++)
+Line 4296 
 for (;; ptr++)
        if (!possessive_quantifier &&
            repeat_max < 0 &&
-           check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
+           check_auto_possessive(previous, utf8, ptr + 1, options, cd))
          {
          repeat_type = 0;    /* Force greedy */
          possessive_quantifier = TRUE;
-Line 4213 
 for (;; ptr++)
+Line 4465 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
               *previous == OP_XCLASS ||
  #endif
-              *previous == OP_REF)
+              *previous == OP_REF ||
+              *previous == OP_REFI)
        {
        if (repeat_max == 0)
          {
-Line 4247 
 for (;; ptr++)
+Line 4500 
 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" BRA and
+     KET opcodes, as this is the place where they get converted into the more
+     special varieties. */
      else if (*previous == OP_BRA  || *previous == OP_CBRA ||
               *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 */
        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
-Line 4265 
 for (;; ptr++)
+Line 4520 
 for (;; ptr++)
          goto FAILED;
          }
-       /* If the maximum repeat count is unlimited, find the end of the bracket
-       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
-       the final KET, so we can't find the end just by going back from the code
-       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
        OP_BRAZERO in front of it, and because the group appears once in the
        data, whereas in other cases it appears the minimum number of times. For
-Line 4319 
 for (;; ptr++)
+Line 4561 
 for (;; ptr++)
              *previous++ = OP_SKIPZERO;
              goto END_REPEAT;
              }
+           brazeroptr = previous;    /* Save for possessive optimizing */
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 4483 
 for (;; ptr++)
+Line 4726 
 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.
-       don't know if there's been an options resetting after the ket. The
-       correct offset was computed above.
+       (To be done next, after recursion adjusted)
+       However, possessively repeated
+       ONCE brackets can be converted into non-capturing brackets, as the
+       behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
+       deal with possessive ONCEs specially.
+       (....)
+       Otherwise, if the quantifier was possessive, 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 cancel the possessive flag so that
+       the default action below, of wrapping everything inside atomic brackets,
+       does not happen.
        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,
+       this group is one that could match an empty string. If so, convert the
-       convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
+       initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
-       that runtime checking can be done. [This check is also applied to
+       checking can be done. [This check is also applied to ONCE groups at
-       atomic groups at runtime, but in a different way.] */
+       runtime, but in a different way.] */
        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)
+ /****
+         if (*bracode == OP_ONCE && possessive_quantifier)
+           *bracode = OP_BRA;
+ ****/
+         if (*bracode == OP_ONCE)
+           *ketcode = OP_KETRMAX + repeat_type;
+         else
            {
-           uschar *scode = bracode;
+           if (possessive_quantifier)
-           do
              {
-             if (could_be_empty_branch(scode, ketcode, utf8, cd))
+             *bracode += 1;                   /* Switch to xxxPOS opcodes */
+             *ketcode = OP_KETRPOS;
+             if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
+             possessive_quantifier = FALSE;
+             }
+           else *ketcode = OP_KETRMAX + repeat_type;
+           if (lengthptr == NULL)
+             {
+             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);
            }
          }
        }
-Line 4532 
 for (;; ptr++)
+Line 4808 
 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.
+     Possessively repeated subpatterns have already been handled in the code
+     just above, so possessive_quantifier is always FALSE for them 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 4569 
 for (;; ptr++)
+Line 4850 
 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 4637 
 for (;; ptr++)
+Line 4928 
 for (;; ptr++)
          arg = ++ptr;
          while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
            || *ptr == '_') ptr++;
-         arglen = ptr - arg;
+         arglen = (int)(ptr - arg);
          }
        if (*ptr != CHAR_RIGHT_PARENTHESIS)
-Line 4653 
 for (;; ptr++)
+Line 4944 
 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;
              }
-           /* 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 */
                {
                *errorcodeptr = ERR66;
                goto FAILED;
                }
-             *code++ = verbs[i].op;
+             *code = verbs[i].op;
+             if (*code++ == OP_THEN)
+               {
+               PUT(code, 0, code - bcptr->current_branch - 1);
+               code += LINK_SIZE;
+               }
              }
            else
-Line 4685 
 for (;; ptr++)
+Line 4988 
 for (;; ptr++)
                *errorcodeptr = ERR59;
                goto FAILED;
                }
-             *code++ = verbs[i].op_arg;
+             *code = verbs[i].op_arg;
+             if (*code++ == OP_THEN_ARG)
+               {
+               PUT(code, 0, code - bcptr->current_branch - 1);
+               code += LINK_SIZE;
+               }
              *code++ = arglen;
              memcpy(code, arg, arglen);
              code += arglen;
-Line 4879 
 for (;; ptr++)
+Line 5187 
 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 4947 
 for (;; ptr++)
+Line 5255 
 for (;; ptr++)
          /* ------------------------------------------------------------ */
          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
          bravalue = OP_ASSERT;
+         cd->assert_depth += 1;
          ptr++;
          break;
-Line 4961 
 for (;; ptr++)
+Line 5270 
 for (;; ptr++)
            continue;
            }
          bravalue = OP_ASSERT_NOT;
+         cd->assert_depth += 1;
          break;
-Line 4970 
 for (;; ptr++)
+Line 5280 
 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 5180 
 for (;; ptr++)
+Line 5492 
 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 5200 
 for (;; ptr++)
+Line 5518 
 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 5225 
 for (;; ptr++)
+Line 5558 
 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 5336 
 for (;; ptr++)
+Line 5669 
 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 5365 
 for (;; ptr++)
+Line 5698 
 for (;; ptr++)
            /* Insert the recursion/subroutine item, automatically wrapped inside
            "once" brackets. Set up a "previous group" length so that a
            subsequent quantifier will work. */
            *code = OP_ONCE;
            PUT(code, 1, 2 + 2*LINK_SIZE);
            code += 1 + LINK_SIZE;
-Line 5439 
 for (;; ptr++)
+Line 5772 
 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 5452 
 for (;; ptr++)
+Line 5784 
 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 5503 
 for (;; ptr++)
+Line 5828 
 for (;; ptr++)
      /* Process nested bracketed regex. Assertions may not be repeated, but
      other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
-     non-register variable in order to be able to pass its address because some
+     non-register variable (tempcode) in order to be able to pass its address
-     compilers complain otherwise. Pass in a new setting for the ims options if
+     because some compilers complain otherwise. */
-     they have changed. */
      previous = (bravalue >= OP_ONCE)? code : NULL;
      *code = bravalue;
-Line 5515 
 for (;; ptr++)
+Line 5839 
 for (;; ptr++)
      if (!compile_regex(
           newoptions,                   /* The complete new option state */
-          options & PCRE_IMS,           /* The previous ims option state */
           &tempcode,                    /* Where to put code (updated) */
           &ptr,                         /* Input pointer (updated) */
           errorcodeptr,                 /* Where to put an error message */
-Line 5531 
 for (;; ptr++)
+Line 5854 
 for (;; ptr++)
             &length_prevgroup           /* Pre-compile phase */
           ))
        goto FAILED;
+     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
-Line 5602 
 for (;; ptr++)
+Line 5928 
 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 5675 
 for (;; ptr++)
+Line 6001 
 for (;; ptr++)
      /* ===================================================================*/
      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
      are arranged to be the negation of the corresponding OP_values in the
      default case when PCRE_UCP is not set. For the back references, the values
      are ESC_REF plus the reference number. Only back references and those types
      that consume a character may be repeated. We can test for values between
-Line 5794 
 for (;; ptr++)
+Line 6120 
 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 5853 
 for (;; ptr++)
+Line 6179 
 for (;; ptr++)
            ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
            }
          else
  #endif
            {
            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
            *code++ = -c;
            }
          }
        continue;
        }
-Line 5902 
 for (;; ptr++)
+Line 6228 
 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 5966 
 return FALSE;
+Line 6292 
 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
-Line 5995 
 Returns:         TRUE on success
+Line 6315 
 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 *lengthptr)
-Line 6012 
 int branchfirstbyte, branchreqbyte;
+Line 6332 
 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 6036 
 pre-compile phase to find out whether an
+Line 6355 
 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 6062 
 for (;;)
+Line 6383 
 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 6091 
 for (;;)
+Line 6403 
 for (;;)
      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 6160 
 for (;;)
+Line 6463 
 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 6181 
 for (;;)
+Line 6485 
 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 6227 
 for (;;)
+Line 6529 
 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 6295 
 for (;;)
+Line 6588 
 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 6317 
 of the more common cases more precisely.
+Line 6610 
 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 6327 
 Returns:     TRUE or FALSE
+Line 6619 
 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)
       {
-      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 6370 
 do {
+Line 6664 
 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 6408 
 is_startline(const uschar *code, unsigne
+Line 6700 
 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 6435 
 do {
+Line 6727 
 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 6472 
 do {
+Line 6766 
 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 6498 
 we return that char, otherwise -1.
+Line 6792 
 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 6520 
 do {
+Line 6814 
 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_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 6666 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
+Line 6973 
 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 6689 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
+Line 6998 
 while (ptr[skipatstart] == CHAR_LEFT_PAR
      options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
    else break;
    }
  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 6715 
 if (utf8)
+Line 7027 
 if (utf8)
  if ((options & PCRE_UCP) != 0)
    {
    errorcode = ERR67;
    goto PCRE_EARLY_ERROR_RETURN;
    }
  #endif
  /* Check validity of \R options. */
-Line 6812 
 outside can help speed up starting point
+Line 7124 
 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, &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 6868 
 field; this time it's used for rememberi
+Line 7179 
 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;
-Line 6886 
 of the function here. */
+Line 7198 
 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,
-   &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;
-Line 6953 
 if (cd->check_lookbehind)
+Line 7265 
 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)
-Line 6992 
 start with ^. and also when all branches
+Line 7305 
 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.530
 


changed lines


 
Added in v.618
 Legend:



Removed from v.530
 


changed lines


 
Added in v.618
-Removed from v.530
+Added in v.618

webmaster@exim.org	ViewVC Help
Powered by ViewVC 1.1.12