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

Diff of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 574 by ph10, Sat Nov 20 17:47:27 2010 UTC revision 779 by ph10, Fri Dec 2 10:39:32 2011 UTC
# Line 6  Line 6 
6  and semantics are as close as possible to those of the Perl 5 language.  and semantics are as close as possible to those of the Perl 5 language.
7    
8                         Written by Philip Hazel                         Written by Philip Hazel
9             Copyright (c) 1997-2010 University of Cambridge             Copyright (c) 1997-2011 University of Cambridge
10    
11  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
12  Redistribution and use in source and binary forms, with or without  Redistribution and use in source and binary forms, with or without
# Line 88  so this number is very generous. Line 88  so this number is very generous.
88  The same workspace is used during the second, actual compile phase for  The same workspace is used during the second, actual compile phase for
89  remembering forward references to groups so that they can be filled in at the  remembering forward references to groups so that they can be filled in at the
90  end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE  end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91  is 4 there is plenty of room. */  is 4 there is plenty of room for most patterns. However, the memory can get
92    filled up by repetitions of forward references, for example patterns like
93    /(?1){0,1999}(b)/, and one user did hit the limit. The code has been changed so
94    that the workspace is expanded using malloc() in this situation. The value
95    below is therefore a minimum, and we put a maximum on it for safety. The
96    minimum is now also defined in terms of LINK_SIZE so that the use of malloc()
97    kicks in at the same number of forward references in all cases. */
98    
99  #define COMPILE_WORK_SIZE (4096)  #define COMPILE_WORK_SIZE (2048*LINK_SIZE)
100    #define COMPILE_WORK_SIZE_MAX (100*COMPILE_WORK_SIZE)
101    
102  /* The overrun tests check for a slightly smaller size so that they detect the  /* The overrun tests check for a slightly smaller size so that they detect the
103  overrun before it actually does run off the end of the data block. */  overrun before it actually does run off the end of the data block. */
104    
105  #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)  #define WORK_SIZE_SAFETY_MARGIN (100)
106    
107    
108  /* Table for handling escaped characters in the range '0'-'z'. Positive returns  /* 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[] =
400    "internal error: previously-checked referenced subpattern not found\0"    "internal error: previously-checked referenced subpattern not found\0"
401    "DEFINE group contains more than one branch\0"    "DEFINE group contains more than one branch\0"
402    /* 55 */    /* 55 */
403    "repeating a DEFINE group is not allowed\0"    "repeating a DEFINE group is not allowed\0"  /** DEAD **/
404    "inconsistent NEWLINE options\0"    "inconsistent NEWLINE options\0"
405    "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"    "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
406    "a numbered reference must not be zero\0"    "a numbered reference must not be zero\0"
# Line 408  static const char error_texts[] = Line 415  static const char error_texts[] =
415    "different names for subpatterns of the same number are not allowed\0"    "different names for subpatterns of the same number are not allowed\0"
416    "(*MARK) must have an argument\0"    "(*MARK) must have an argument\0"
417    "this version of PCRE is not compiled with PCRE_UCP support\0"    "this version of PCRE is not compiled with PCRE_UCP support\0"
418    "\\c must be followed by an ASCII character\0"    "\\c must be followed by an ASCII character\0"
419      "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
420      /* 70 */
421      "internal error: unknown opcode in find_fixedlength()\0"
422      "\\N is not supported in a class\0"
423      "too many forward references\0"
424    ;    ;
425    
426  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
# Line 545  static const unsigned char ebcdic_charta Line 557  static const unsigned char ebcdic_charta
557  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
558    
559  static BOOL  static BOOL
560    compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
561      int *, int *, branch_chain *, compile_data *, int *);      int *, int *, branch_chain *, compile_data *, int *);
562    
563    
# Line 577  return s; Line 589  return s;
589    
590    
591  /*************************************************  /*************************************************
592    *           Expand the workspace                 *
593    *************************************************/
594    
595    /* This function is called during the second compiling phase, if the number of
596    forward references fills the existing workspace, which is originally a block on
597    the stack. A larger block is obtained from malloc() unless the ultimate limit
598    has been reached or the increase will be rather small.
599    
600    Argument: pointer to the compile data block
601    Returns:  0 if all went well, else an error number
602    */
603    
604    static int
605    expand_workspace(compile_data *cd)
606    {
607    uschar *newspace;
608    int newsize = cd->workspace_size * 2;
609    
610    if (newsize > COMPILE_WORK_SIZE_MAX) newsize = COMPILE_WORK_SIZE_MAX;
611    if (cd->workspace_size >= COMPILE_WORK_SIZE_MAX ||
612        newsize - cd->workspace_size < WORK_SIZE_SAFETY_MARGIN)
613     return ERR72;
614    
615    newspace = (pcre_malloc)(newsize);
616    if (newspace == NULL) return ERR21;
617    
618    memcpy(newspace, cd->start_workspace, cd->workspace_size);
619    cd->hwm = (uschar *)newspace + (cd->hwm - cd->start_workspace);
620    if (cd->workspace_size > COMPILE_WORK_SIZE)
621      (pcre_free)((void *)cd->start_workspace);
622    cd->start_workspace = newspace;
623    cd->workspace_size = newsize;
624    return 0;
625    }
626    
627    
628    
629    /*************************************************
630    *            Check for counted repeat            *
631    *************************************************/
632    
633    /* This function is called when a '{' is encountered in a place where it might
634    start a quantifier. It looks ahead to see if it really is a quantifier or not.
635    It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
636    where the ddds are digits.
637    
638    Arguments:
639      p         pointer to the first char after '{'
640    
641    Returns:    TRUE or FALSE
642    */
643    
644    static BOOL
645    is_counted_repeat(const uschar *p)
646    {
647    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
648    while ((digitab[*p] & ctype_digit) != 0) p++;
649    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
650    
651    if (*p++ != CHAR_COMMA) return FALSE;
652    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
653    
654    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
655    while ((digitab[*p] & ctype_digit) != 0) p++;
656    
657    return (*p == CHAR_RIGHT_CURLY_BRACKET);
658    }
659    
660    
661    
662    /*************************************************
663  *            Handle escapes                      *  *            Handle escapes                      *
664  *************************************************/  *************************************************/
665    
# Line 642  else Line 725  else
725    
726      case CHAR_l:      case CHAR_l:
727      case CHAR_L:      case CHAR_L:
728        *errorcodeptr = ERR37;
729        break;
730    
731      case CHAR_u:      case CHAR_u:
732        if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
733          {
734          /* In JavaScript, \u must be followed by four hexadecimal numbers.
735          Otherwise it is a lowercase u letter. */
736          if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
737               && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
738            {
739            c = 0;
740            for (i = 0; i < 4; ++i)
741              {
742              register int cc = *(++ptr);
743    #ifndef EBCDIC  /* ASCII/UTF-8 coding */
744              if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
745              c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
746    #else           /* EBCDIC coding */
747              if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
748              c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
749    #endif
750              }
751            }
752          }
753        else
754          *errorcodeptr = ERR37;
755        break;
756    
757      case CHAR_U:      case CHAR_U:
758      *errorcodeptr = ERR37;      /* In JavaScript, \U is an uppercase U letter. */
759        if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
760      break;      break;
761    
762      /* \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
763        class, \g must be followed by one of a number of specific things:
764    
765      (1) A number, either plain or braced. If positive, it is an absolute      (1) A number, either plain or braced. If positive, it is an absolute
766      backreference. If negative, it is a relative backreference. This is a Perl      backreference. If negative, it is a relative backreference. This is a Perl
# Line 664  else Line 777  else
777      the -ESC_g code (cf \k). */      the -ESC_g code (cf \k). */
778    
779      case CHAR_g:      case CHAR_g:
780        if (isclass) break;
781      if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)      if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
782        {        {
783        c = -ESC_g;        c = -ESC_g;
# Line 792  else Line 906  else
906      treated as a data character. */      treated as a data character. */
907    
908      case CHAR_x:      case CHAR_x:
909        if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
910          {
911          /* In JavaScript, \x must be followed by two hexadecimal numbers.
912          Otherwise it is a lowercase x letter. */
913          if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
914            {
915            c = 0;
916            for (i = 0; i < 2; ++i)
917              {
918              register int cc = *(++ptr);
919    #ifndef EBCDIC  /* ASCII/UTF-8 coding */
920              if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
921              c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
922    #else           /* EBCDIC coding */
923              if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
924              c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
925    #endif
926              }
927            }
928          break;
929          }
930    
931      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
932        {        {
933        const uschar *pt = ptr + 2;        const uschar *pt = ptr + 2;
# Line 857  else Line 993  else
993      if (c > 127)  /* Excludes all non-ASCII in either mode */      if (c > 127)  /* Excludes all non-ASCII in either mode */
994        {        {
995        *errorcodeptr = ERR68;        *errorcodeptr = ERR68;
996        break;        break;
997        }        }
998      if (c >= CHAR_a && c <= CHAR_z) c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
999      c ^= 0x40;      c ^= 0x40;
1000  #else             /* EBCDIC coding */  #else             /* EBCDIC coding */
# Line 885  else Line 1021  else
1021    }    }
1022    
1023  /* Perl supports \N{name} for character names, as well as plain \N for "not  /* Perl supports \N{name} for character names, as well as plain \N for "not
1024  newline". PCRE does not support \N{name}. */  newline". PCRE does not support \N{name}. However, it does support
1025    quantification such as \N{2,3}. */
1026    
1027  if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)  if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
1028         !is_counted_repeat(ptr+2))
1029    *errorcodeptr = ERR37;    *errorcodeptr = ERR37;
1030    
1031  /* If PCRE_UCP is set, we change the values for \d etc. */  /* If PCRE_UCP is set, we change the values for \d etc. */
# Line 997  return -1; Line 1135  return -1;
1135    
1136    
1137  /*************************************************  /*************************************************
 *            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);  
 }  
   
   
   
 /*************************************************  
1138  *         Read repeat counts                     *  *         Read repeat counts                     *
1139  *************************************************/  *************************************************/
1140    
# Line 1105  top-level call starts at the beginning o Line 1210  top-level call starts at the beginning o
1210  start at a parenthesis. It scans along a pattern's text looking for capturing  start at a parenthesis. It scans along a pattern's text looking for capturing
1211  subpatterns, and counting them. If it finds a named pattern that matches the  subpatterns, and counting them. If it finds a named pattern that matches the
1212  name it is given, it returns its number. Alternatively, if the name is NULL, it  name it is given, it returns its number. Alternatively, if the name is NULL, it
1213  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
1214  encountered, the name will be terminated by '>' because that is checked in the  track of subpatterns that reset the capturing group numbers - the (?| feature.
1215  first pass. Recursion is used to keep track of subpatterns that reset the  
1216  capturing group numbers - the (?| feature.  This function was originally called only from the second pass, in which we know
1217    that if (?< or (?' or (?P< is encountered, the name will be correctly
1218    terminated because that is checked in the first pass. There is now one call to
1219    this function in the first pass, to check for a recursive back reference by
1220    name (so that we can make the whole group atomic). In this case, we need check
1221    only up to the current position in the pattern, and that is still OK because
1222    and previous occurrences will have been checked. To make this work, the test
1223    for "end of pattern" is a check against cd->end_pattern in the main loop,
1224    instead of looking for a binary zero. This means that the special first-pass
1225    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1226    processing items within the loop are OK, because afterwards the main loop will
1227    terminate.)
1228    
1229  Arguments:  Arguments:
1230    ptrptr       address of the current character pointer (updated)    ptrptr       address of the current character pointer (updated)
# Line 1116  Arguments: Line 1232  Arguments:
1232    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1233    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1234    xmode        TRUE if we are in /x mode    xmode        TRUE if we are in /x mode
1235    utf8         TRUE if we are in UTF-8 mode    utf8         TRUE if we are in UTF-8 mode
1236    count        pointer to the current capturing subpattern number (updated)    count        pointer to the current capturing subpattern number (updated)
1237    
1238  Returns:       the number of the named subpattern, or -1 if not found  Returns:       the number of the named subpattern, or -1 if not found
# Line 1209  if (ptr[0] == CHAR_LEFT_PARENTHESIS) Line 1325  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1325    }    }
1326    
1327  /* Past any initial parenthesis handling, scan for parentheses or vertical  /* Past any initial parenthesis handling, scan for parentheses or vertical
1328  bars. */  bars. Stop if we get to cd->end_pattern. Note that this is important for the
1329    first-pass call when this value is temporarily adjusted to stop at the current
1330    position. So DO NOT change this to a test for binary zero. */
1331    
1332  for (; *ptr != 0; ptr++)  for (; ptr < cd->end_pattern; ptr++)
1333    {    {
1334    /* Skip over backslashed characters and also entire \Q...\E */    /* Skip over backslashed characters and also entire \Q...\E */
1335    
# Line 1285  for (; *ptr != 0; ptr++) Line 1403  for (; *ptr != 0; ptr++)
1403    
1404    if (xmode && *ptr == CHAR_NUMBER_SIGN)    if (xmode && *ptr == CHAR_NUMBER_SIGN)
1405      {      {
1406      ptr++;      ptr++;
1407      while (*ptr != 0)      while (*ptr != 0)
1408        {        {
1409        if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }        if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1410        ptr++;        ptr++;
1411  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1412        if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;        if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1413  #endif  #endif
1414        }        }
# Line 1348  Arguments: Line 1466  Arguments:
1466    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1467    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1468    xmode        TRUE if we are in /x mode    xmode        TRUE if we are in /x mode
1469    utf8         TRUE if we are in UTF-8 mode    utf8         TRUE if we are in UTF-8 mode
1470    
1471  Returns:       the number of the found subpattern, or -1 if not found  Returns:       the number of the found subpattern, or -1 if not found
1472  */  */
# Line 1384  return rc; Line 1502  return rc;
1502    
1503  /* This is called by several functions that scan a compiled expression looking  /* This is called by several functions that scan a compiled expression looking
1504  for a fixed first character, or an anchoring op code etc. It skips over things  for a fixed first character, or an anchoring op code etc. It skips over things
1505  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
1506  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
1507  assertions, and also the \b assertion; for others it does not.  does not.
1508    
1509  Arguments:  Arguments:
1510    code         pointer to the start of the group    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  
1511    skipassert   TRUE if certain assertions are to be skipped    skipassert   TRUE if certain assertions are to be skipped
1512    
1513  Returns:       pointer to the first significant opcode  Returns:       pointer to the first significant opcode
1514  */  */
1515    
1516  static const uschar*  static const uschar*
1517  first_significant_code(const uschar *code, int *options, int optbit,  first_significant_code(const uschar *code, BOOL skipassert)
   BOOL skipassert)  
1518  {  {
1519  for (;;)  for (;;)
1520    {    {
1521    switch ((int)*code)    switch ((int)*code)
1522      {      {
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
   
1523      case OP_ASSERT_NOT:      case OP_ASSERT_NOT:
1524      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1525      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
# Line 1461  and doing the check at the end; a flag s Line 1569  and doing the check at the end; a flag s
1569    
1570  Arguments:  Arguments:
1571    code     points to the start of the pattern (the bracket)    code     points to the start of the pattern (the bracket)
1572    options  the compiling options    utf8     TRUE in UTF-8 mode
1573    atend    TRUE if called when the pattern is complete    atend    TRUE if called when the pattern is complete
1574    cd       the "compile data" structure    cd       the "compile data" structure
1575    
1576  Returns:   the fixed length,  Returns:   the fixed length,
1577               or -1 if there is no fixed length,               or -1 if there is no fixed length,
1578               or -2 if \C was encountered               or -2 if \C was encountered (in UTF-8 mode only)
1579               or -3 if an OP_RECURSE item was encountered and atend is FALSE               or -3 if an OP_RECURSE item was encountered and atend is FALSE
1580                 or -4 if an unknown opcode was encountered (internal error)
1581  */  */
1582    
1583  static int  static int
1584  find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1585  {  {
1586  int length = -1;  int length = -1;
1587    
# Line 1489  for (;;) Line 1598  for (;;)
1598    register int op = *cc;    register int op = *cc;
1599    switch (op)    switch (op)
1600      {      {
1601        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1602        OP_BRA (normal non-capturing bracket) because the other variants of these
1603        opcodes are all concerned with unlimited repeated groups, which of course
1604        are not of fixed length. */
1605    
1606      case OP_CBRA:      case OP_CBRA:
1607      case OP_BRA:      case OP_BRA:
1608      case OP_ONCE:      case OP_ONCE:
1609        case OP_ONCE_NC:
1610      case OP_COND:      case OP_COND:
1611      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1612      if (d < 0) return d;      if (d < 0) return d;
1613      branchlength += d;      branchlength += d;
1614      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
1615      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1616      break;      break;
1617    
1618      /* 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.
1619      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
1620      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
1621        the same code. Note that we must not include the OP_KETRxxx opcodes here,
1622        because they all imply an unlimited repeat. */
1623    
1624      case OP_ALT:      case OP_ALT:
1625      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1626      case OP_END:      case OP_END:
1627        case OP_ACCEPT:
1628        case OP_ASSERT_ACCEPT:
1629      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1630        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
1631      if (*cc != OP_ALT) return length;      if (*cc != OP_ALT) return length;
# Line 1525  for (;;) Line 1642  for (;;)
1642      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1643      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1644      if (cc > cs && cc < ce) return -1;                /* Recursion */      if (cc > cs && cc < ce) return -1;                /* Recursion */
1645      d = find_fixedlength(cs + 2, options, atend, cd);      d = find_fixedlength(cs + 2, utf8, atend, cd);
1646      if (d < 0) return d;      if (d < 0) return d;
1647      branchlength += d;      branchlength += d;
1648      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
# Line 1542  for (;;) Line 1659  for (;;)
1659    
1660      /* Skip over things that don't match chars */      /* Skip over things that don't match chars */
1661    
1662      case OP_REVERSE:      case OP_MARK:
1663        case OP_PRUNE_ARG:
1664        case OP_SKIP_ARG:
1665        case OP_THEN_ARG:
1666        cc += cc[1] + _pcre_OP_lengths[*cc];
1667        break;
1668    
1669        case OP_CALLOUT:
1670        case OP_CIRC:
1671        case OP_CIRCM:
1672        case OP_CLOSE:
1673        case OP_COMMIT:
1674      case OP_CREF:      case OP_CREF:
     case OP_NCREF:  
     case OP_RREF:  
     case OP_NRREF:  
1675      case OP_DEF:      case OP_DEF:
1676      case OP_OPT:      case OP_DOLL:
1677      case OP_CALLOUT:      case OP_DOLLM:
     case OP_SOD:  
     case OP_SOM:  
     case OP_SET_SOM:  
1678      case OP_EOD:      case OP_EOD:
1679      case OP_EODN:      case OP_EODN:
1680      case OP_CIRC:      case OP_FAIL:
1681      case OP_DOLL:      case OP_NCREF:
1682        case OP_NRREF:
1683      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1684        case OP_PRUNE:
1685        case OP_REVERSE:
1686        case OP_RREF:
1687        case OP_SET_SOM:
1688        case OP_SKIP:
1689        case OP_SOD:
1690        case OP_SOM:
1691        case OP_THEN:
1692      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1693      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
1694      break;      break;
# Line 1565  for (;;) Line 1696  for (;;)
1696      /* Handle literal characters */      /* Handle literal characters */
1697    
1698      case OP_CHAR:      case OP_CHAR:
1699      case OP_CHARNC:      case OP_CHARI:
1700      case OP_NOT:      case OP_NOT:
1701        case OP_NOTI:
1702      branchlength++;      branchlength++;
1703      cc += 2;      cc += 2;
1704  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1705      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];  
1706  #endif  #endif
1707      break;      break;
1708    
# Line 1579  for (;;) Line 1710  for (;;)
1710      need to skip over a multibyte character in UTF8 mode.  */      need to skip over a multibyte character in UTF8 mode.  */
1711    
1712      case OP_EXACT:      case OP_EXACT:
1713        case OP_EXACTI:
1714        case OP_NOTEXACT:
1715        case OP_NOTEXACTI:
1716      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1717      cc += 4;      cc += 4;
1718  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1719      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];  
1720  #endif  #endif
1721      break;      break;
1722    
# Line 1600  for (;;) Line 1733  for (;;)
1733      cc += 2;      cc += 2;
1734      /* Fall through */      /* Fall through */
1735    
1736        case OP_HSPACE:
1737        case OP_VSPACE:
1738        case OP_NOT_HSPACE:
1739        case OP_NOT_VSPACE:
1740      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
1741      case OP_DIGIT:      case OP_DIGIT:
1742      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
# Line 1612  for (;;) Line 1749  for (;;)
1749      cc++;      cc++;
1750      break;      break;
1751    
1752      /* The single-byte matcher isn't allowed */      /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1753        otherwise \C is coded as OP_ALLANY. */
1754    
1755      case OP_ANYBYTE:      case OP_ANYBYTE:
1756      return -2;      return -2;
# Line 1631  for (;;) Line 1769  for (;;)
1769    
1770      switch (*cc)      switch (*cc)
1771        {        {
1772          case OP_CRPLUS:
1773          case OP_CRMINPLUS:
1774        case OP_CRSTAR:        case OP_CRSTAR:
1775        case OP_CRMINSTAR:        case OP_CRMINSTAR:
1776        case OP_CRQUERY:        case OP_CRQUERY:
# Line 1651  for (;;) Line 1791  for (;;)
1791    
1792      /* Anything else is variable length */      /* Anything else is variable length */
1793    
1794      default:      case OP_ANYNL:
1795        case OP_BRAMINZERO:
1796        case OP_BRAPOS:
1797        case OP_BRAPOSZERO:
1798        case OP_BRAZERO:
1799        case OP_CBRAPOS:
1800        case OP_EXTUNI:
1801        case OP_KETRMAX:
1802        case OP_KETRMIN:
1803        case OP_KETRPOS:
1804        case OP_MINPLUS:
1805        case OP_MINPLUSI:
1806        case OP_MINQUERY:
1807        case OP_MINQUERYI:
1808        case OP_MINSTAR:
1809        case OP_MINSTARI:
1810        case OP_MINUPTO:
1811        case OP_MINUPTOI:
1812        case OP_NOTMINPLUS:
1813        case OP_NOTMINPLUSI:
1814        case OP_NOTMINQUERY:
1815        case OP_NOTMINQUERYI:
1816        case OP_NOTMINSTAR:
1817        case OP_NOTMINSTARI:
1818        case OP_NOTMINUPTO:
1819        case OP_NOTMINUPTOI:
1820        case OP_NOTPLUS:
1821        case OP_NOTPLUSI:
1822        case OP_NOTPOSPLUS:
1823        case OP_NOTPOSPLUSI:
1824        case OP_NOTPOSQUERY:
1825        case OP_NOTPOSQUERYI:
1826        case OP_NOTPOSSTAR:
1827        case OP_NOTPOSSTARI:
1828        case OP_NOTPOSUPTO:
1829        case OP_NOTPOSUPTOI:
1830        case OP_NOTQUERY:
1831        case OP_NOTQUERYI:
1832        case OP_NOTSTAR:
1833        case OP_NOTSTARI:
1834        case OP_NOTUPTO:
1835        case OP_NOTUPTOI:
1836        case OP_PLUS:
1837        case OP_PLUSI:
1838        case OP_POSPLUS:
1839        case OP_POSPLUSI:
1840        case OP_POSQUERY:
1841        case OP_POSQUERYI:
1842        case OP_POSSTAR:
1843        case OP_POSSTARI:
1844        case OP_POSUPTO:
1845        case OP_POSUPTOI:
1846        case OP_QUERY:
1847        case OP_QUERYI:
1848        case OP_REF:
1849        case OP_REFI:
1850        case OP_SBRA:
1851        case OP_SBRAPOS:
1852        case OP_SCBRA:
1853        case OP_SCBRAPOS:
1854        case OP_SCOND:
1855        case OP_SKIPZERO:
1856        case OP_STAR:
1857        case OP_STARI:
1858        case OP_TYPEMINPLUS:
1859        case OP_TYPEMINQUERY:
1860        case OP_TYPEMINSTAR:
1861        case OP_TYPEMINUPTO:
1862        case OP_TYPEPLUS:
1863        case OP_TYPEPOSPLUS:
1864        case OP_TYPEPOSQUERY:
1865        case OP_TYPEPOSSTAR:
1866        case OP_TYPEPOSUPTO:
1867        case OP_TYPEQUERY:
1868        case OP_TYPESTAR:
1869        case OP_TYPEUPTO:
1870        case OP_UPTO:
1871        case OP_UPTOI:
1872      return -1;      return -1;
1873    
1874        /* Catch unrecognized opcodes so that when new ones are added they
1875        are not forgotten, as has happened in the past. */
1876    
1877        default:
1878        return -4;
1879      }      }
1880    }    }
1881  /* Control never gets here */  /* Control never gets here */
# Line 1685  _pcre_find_bracket(const uschar *code, B Line 1908  _pcre_find_bracket(const uschar *code, B
1908  for (;;)  for (;;)
1909    {    {
1910    register int c = *code;    register int c = *code;
1911    
1912    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1913    
1914    /* XCLASS is used for classes that cannot be represented just by a bit    /* XCLASS is used for classes that cannot be represented just by a bit
# Line 1703  for (;;) Line 1927  for (;;)
1927    
1928    /* Handle capturing bracket */    /* Handle capturing bracket */
1929    
1930    else if (c == OP_CBRA)    else if (c == OP_CBRA || c == OP_SCBRA ||
1931               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1932      {      {
1933      int n = GET2(code, 1+LINK_SIZE);      int n = GET2(code, 1+LINK_SIZE);
1934      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
# Line 1745  for (;;) Line 1970  for (;;)
1970        break;        break;
1971    
1972        case OP_THEN_ARG:        case OP_THEN_ARG:
1973        code += code[1+LINK_SIZE];        code += code[1];
1974        break;        break;
1975        }        }
1976    
# Line 1761  for (;;) Line 1986  for (;;)
1986      if (utf8) switch(c)      if (utf8) switch(c)
1987        {        {
1988        case OP_CHAR:        case OP_CHAR:
1989        case OP_CHARNC:        case OP_CHARI:
1990        case OP_EXACT:        case OP_EXACT:
1991          case OP_EXACTI:
1992        case OP_UPTO:        case OP_UPTO:
1993          case OP_UPTOI:
1994        case OP_MINUPTO:        case OP_MINUPTO:
1995          case OP_MINUPTOI:
1996        case OP_POSUPTO:        case OP_POSUPTO:
1997          case OP_POSUPTOI:
1998        case OP_STAR:        case OP_STAR:
1999          case OP_STARI:
2000        case OP_MINSTAR:        case OP_MINSTAR:
2001          case OP_MINSTARI:
2002        case OP_POSSTAR:        case OP_POSSTAR:
2003          case OP_POSSTARI:
2004        case OP_PLUS:        case OP_PLUS:
2005          case OP_PLUSI:
2006        case OP_MINPLUS:        case OP_MINPLUS:
2007          case OP_MINPLUSI:
2008        case OP_POSPLUS:        case OP_POSPLUS:
2009          case OP_POSPLUSI:
2010        case OP_QUERY:        case OP_QUERY:
2011          case OP_QUERYI:
2012        case OP_MINQUERY:        case OP_MINQUERY:
2013          case OP_MINQUERYI:
2014        case OP_POSQUERY:        case OP_POSQUERY:
2015          case OP_POSQUERYI:
2016        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2017        break;        break;
2018        }        }
# Line 1851  for (;;) Line 2089  for (;;)
2089        break;        break;
2090    
2091        case OP_THEN_ARG:        case OP_THEN_ARG:
2092        code += code[1+LINK_SIZE];        code += code[1];
2093        break;        break;
2094        }        }
2095    
# Line 1867  for (;;) Line 2105  for (;;)
2105      if (utf8) switch(c)      if (utf8) switch(c)
2106        {        {
2107        case OP_CHAR:        case OP_CHAR:
2108        case OP_CHARNC:        case OP_CHARI:
2109        case OP_EXACT:        case OP_EXACT:
2110          case OP_EXACTI:
2111        case OP_UPTO:        case OP_UPTO:
2112          case OP_UPTOI:
2113        case OP_MINUPTO:        case OP_MINUPTO:
2114          case OP_MINUPTOI:
2115        case OP_POSUPTO:        case OP_POSUPTO:
2116          case OP_POSUPTOI:
2117        case OP_STAR:        case OP_STAR:
2118          case OP_STARI:
2119        case OP_MINSTAR:        case OP_MINSTAR:
2120          case OP_MINSTARI:
2121        case OP_POSSTAR:        case OP_POSSTAR:
2122          case OP_POSSTARI:
2123        case OP_PLUS:        case OP_PLUS:
2124          case OP_PLUSI:
2125        case OP_MINPLUS:        case OP_MINPLUS:
2126          case OP_MINPLUSI:
2127        case OP_POSPLUS:        case OP_POSPLUS:
2128          case OP_POSPLUSI:
2129        case OP_QUERY:        case OP_QUERY:
2130          case OP_QUERYI:
2131        case OP_MINQUERY:        case OP_MINQUERY:
2132          case OP_MINQUERYI:
2133        case OP_POSQUERY:        case OP_POSQUERY:
2134          case OP_POSQUERYI:
2135        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2136        break;        break;
2137        }        }
# Line 1919  could_be_empty_branch(const uschar *code Line 2170  could_be_empty_branch(const uschar *code
2170    compile_data *cd)    compile_data *cd)
2171  {  {
2172  register int c;  register int c;
2173  for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2174       code < endcode;       code < endcode;
2175       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2176    {    {
2177    const uschar *ccode;    const uschar *ccode;
2178    
# Line 1937  for (code = first_significant_code(code Line 2188  for (code = first_significant_code(code
2188      continue;      continue;
2189      }      }
2190    
   /* 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;  
     }  
   
2191    /* For a recursion/subroutine call, if its end has been reached, which    /* For a recursion/subroutine call, if its end has been reached, which
2192    implies a subroutine call, we can scan it. */    implies a backward reference subroutine call, we can scan it. If it's a
2193      forward reference subroutine call, we can't. To detect forward reference
2194      we have to scan up the list that is kept in the workspace. This function is
2195      called only when doing the real compile, not during the pre-compile that
2196      measures the size of the compiled pattern. */
2197    
2198    if (c == OP_RECURSE)    if (c == OP_RECURSE)
2199      {      {
2200      BOOL empty_branch = FALSE;      const uschar *scode;
2201      const uschar *scode = cd->start_code + GET(code, 1);      BOOL empty_branch;
2202    
2203        /* Test for forward reference */
2204    
2205        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2206          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2207    
2208        /* Not a forward reference, test for completed backward reference */
2209    
2210        empty_branch = FALSE;
2211        scode = cd->start_code + GET(code, 1);
2212      if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */      if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2213    
2214        /* Completed backwards reference */
2215    
2216      do      do
2217        {        {
2218        if (could_be_empty_branch(scode, endcode, utf8, cd))        if (could_be_empty_branch(scode, endcode, utf8, cd))
# Line 1965  for (code = first_significant_code(code Line 2223  for (code = first_significant_code(code
2223        scode += GET(scode, 1);        scode += GET(scode, 1);
2224        }        }
2225      while (*scode == OP_ALT);      while (*scode == OP_ALT);
2226    
2227      if (!empty_branch) return FALSE;  /* All branches are non-empty */      if (!empty_branch) return FALSE;  /* All branches are non-empty */
2228      continue;      continue;
2229      }      }
2230    
2231      /* Groups with zero repeats can of course be empty; skip them. */
2232    
2233      if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2234          c == OP_BRAPOSZERO)
2235        {
2236        code += _pcre_OP_lengths[c];
2237        do code += GET(code, 1); while (*code == OP_ALT);
2238        c = *code;
2239        continue;
2240        }
2241    
2242      /* A nested group that is already marked as "could be empty" can just be
2243      skipped. */
2244    
2245      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2246          c == OP_SCBRA || c == OP_SCBRAPOS)
2247        {
2248        do code += GET(code, 1); while (*code == OP_ALT);
2249        c = *code;
2250        continue;
2251        }
2252    
2253    /* For other groups, scan the branches. */    /* For other groups, scan the branches. */
2254    
2255    if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)    if (c == OP_BRA  || c == OP_BRAPOS ||
2256          c == OP_CBRA || c == OP_CBRAPOS ||
2257          c == OP_ONCE || c == OP_ONCE_NC ||
2258          c == OP_COND)
2259      {      {
2260      BOOL empty_branch;      BOOL empty_branch;
2261      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
# Line 2058  for (code = first_significant_code(code Line 2342  for (code = first_significant_code(code
2342      case OP_ALLANY:      case OP_ALLANY:
2343      case OP_ANYBYTE:      case OP_ANYBYTE:
2344      case OP_CHAR:      case OP_CHAR:
2345      case OP_CHARNC:      case OP_CHARI:
2346      case OP_NOT:      case OP_NOT:
2347        case OP_NOTI:
2348      case OP_PLUS:      case OP_PLUS:
2349      case OP_MINPLUS:      case OP_MINPLUS:
2350      case OP_POSPLUS:      case OP_POSPLUS:
# Line 2099  for (code = first_significant_code(code Line 2384  for (code = first_significant_code(code
2384      case OP_KET:      case OP_KET:
2385      case OP_KETRMAX:      case OP_KETRMAX:
2386      case OP_KETRMIN:      case OP_KETRMIN:
2387        case OP_KETRPOS:
2388      case OP_ALT:      case OP_ALT:
2389      return TRUE;      return TRUE;
2390    
# Line 2107  for (code = first_significant_code(code Line 2393  for (code = first_significant_code(code
2393    
2394  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2395      case OP_STAR:      case OP_STAR:
2396        case OP_STARI:
2397      case OP_MINSTAR:      case OP_MINSTAR:
2398        case OP_MINSTARI:
2399      case OP_POSSTAR:      case OP_POSSTAR:
2400        case OP_POSSTARI:
2401      case OP_QUERY:      case OP_QUERY:
2402        case OP_QUERYI:
2403      case OP_MINQUERY:      case OP_MINQUERY:
2404        case OP_MINQUERYI:
2405      case OP_POSQUERY:      case OP_POSQUERY:
2406        case OP_POSQUERYI:
2407      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2408      break;      break;
2409    
2410      case OP_UPTO:      case OP_UPTO:
2411        case OP_UPTOI:
2412      case OP_MINUPTO:      case OP_MINUPTO:
2413        case OP_MINUPTOI:
2414      case OP_POSUPTO:      case OP_POSUPTO:
2415        case OP_POSUPTOI:
2416      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2417      break;      break;
2418  #endif  #endif
# Line 2132  for (code = first_significant_code(code Line 2427  for (code = first_significant_code(code
2427      break;      break;
2428    
2429      case OP_THEN_ARG:      case OP_THEN_ARG:
2430      code += code[1+LINK_SIZE];      code += code[1];
2431      break;      break;
2432    
2433      /* None of the remaining opcodes are required to match a character. */      /* None of the remaining opcodes are required to match a character. */
# Line 2155  return TRUE; Line 2450  return TRUE;
2450  the current branch of the current pattern to see if it could match the empty  the current branch of the current pattern to see if it could match the empty
2451  string. If it could, we must look outwards for branches at other levels,  string. If it could, we must look outwards for branches at other levels,
2452  stopping when we pass beyond the bracket which is the subject of the recursion.  stopping when we pass beyond the bracket which is the subject of the recursion.
2453    This function is called only during the real compile, not during the
2454    pre-compile.
2455    
2456  Arguments:  Arguments:
2457    code        points to start of the recursion    code        points to start of the recursion
# Line 2205  where Perl recognizes it as the POSIX cl Line 2502  where Perl recognizes it as the POSIX cl
2502  "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,  "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2503  I think.  I think.
2504    
2505    A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2506    It seems that the appearance of a nested POSIX class supersedes an apparent
2507    external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2508    a digit.
2509    
2510    In Perl, unescaped square brackets may also appear as part of class names. For
2511    example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2512    [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2513    seem right at all. PCRE does not allow closing square brackets in POSIX class
2514    names.
2515    
2516  Arguments:  Arguments:
2517    ptr      pointer to the initial [    ptr      pointer to the initial [
2518    endptr   where to return the end pointer    endptr   where to return the end pointer
# Line 2219  int terminator;          /* Don't combin Line 2527  int terminator;          /* Don't combin
2527  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2528  for (++ptr; *ptr != 0; ptr++)  for (++ptr; *ptr != 0; ptr++)
2529    {    {
2530    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2531        ptr++;
2532      else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2533      else
2534      {      {
     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;  
2535      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2536        {        {
2537        *endptr = ptr;        *endptr = ptr;
2538        return TRUE;        return TRUE;
2539        }        }
2540        if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2541             (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2542              ptr[1] == CHAR_EQUALS_SIGN) &&
2543            check_posix_syntax(ptr, endptr))
2544          return FALSE;
2545      }      }
2546    }    }
2547  return FALSE;  return FALSE;
# Line 2532  if ((options & PCRE_EXTENDED) != 0) Line 2847  if ((options & PCRE_EXTENDED) != 0)
2847      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2848      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2849        {        {
2850        ptr++;        ptr++;
2851        while (*ptr != 0)        while (*ptr != 0)
2852          {          {
2853          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2854          ptr++;          ptr++;
2855  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2856          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2857  #endif  #endif
2858          }          }
# Line 2576  if ((options & PCRE_EXTENDED) != 0) Line 2891  if ((options & PCRE_EXTENDED) != 0)
2891      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2892      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2893        {        {
2894        ptr++;        ptr++;
2895        while (*ptr != 0)        while (*ptr != 0)
2896          {          {
2897          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2898          ptr++;          ptr++;
2899  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2900          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2901  #endif  #endif
2902          }          }
# Line 2609  if (next >= 0) switch(op_code) Line 2924  if (next >= 0) switch(op_code)
2924  #endif  #endif
2925    return c != next;    return c != next;
2926    
2927    /* 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
2928    Unicode property support, we can use it to test the other case of    Unicode property support, we can use it to test the other case of
2929    high-valued characters. */    high-valued characters. */
2930    
2931    case OP_CHARNC:    case OP_CHARI:
2932  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2933    GETCHARTEST(c, previous);    GETCHARTEST(c, previous);
2934  #else  #else
# Line 2636  if (next >= 0) switch(op_code) Line 2951  if (next >= 0) switch(op_code)
2951  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2952    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2953    
2954    /* 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
2955      opcodes are not used for multi-byte characters, because they are coded using
2956      an XCLASS instead. */
2957    
2958    case OP_NOT:    case OP_NOT:
2959      return (c = *previous) == next;
2960    
2961      case OP_NOTI:
2962    if ((c = *previous) == next) return TRUE;    if ((c = *previous) == next) return TRUE;
   if ((options & PCRE_CASELESS) == 0) return FALSE;  
2963  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2964    if (utf8)    if (utf8)
2965      {      {
# Line 2745  replaced by OP_PROP codes when PCRE_UCP Line 3064  replaced by OP_PROP codes when PCRE_UCP
3064  switch(op_code)  switch(op_code)
3065    {    {
3066    case OP_CHAR:    case OP_CHAR:
3067    case OP_CHARNC:    case OP_CHARI:
3068  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3069    GETCHARTEST(c, previous);    GETCHARTEST(c, previous);
3070  #else  #else
# Line 2932  Arguments: Line 3251  Arguments:
3251    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3252    reqbyteptr     set to the last literal character required, else < 0    reqbyteptr     set to the last literal character required, else < 0
3253    bcptr          points to current branch chain    bcptr          points to current branch chain
3254      cond_depth     conditional nesting depth
3255    cd             contains pointers to tables etc.    cd             contains pointers to tables etc.
3256    lengthptr      NULL during the real compile phase    lengthptr      NULL during the real compile phase
3257                   points to length accumulator during pre-compile phase                   points to length accumulator during pre-compile phase
# Line 2943  Returns:         TRUE on success Line 3263  Returns:         TRUE on success
3263  static BOOL  static BOOL
3264  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3265    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3266    compile_data *cd, int *lengthptr)    int cond_depth, compile_data *cd, int *lengthptr)
3267  {  {
3268  int repeat_type, op_type;  int repeat_type, op_type;
3269  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
# Line 2952  int greedy_default, greedy_non_default; Line 3272  int greedy_default, greedy_non_default;
3272  int firstbyte, reqbyte;  int firstbyte, reqbyte;
3273  int zeroreqbyte, zerofirstbyte;  int zeroreqbyte, zerofirstbyte;
3274  int req_caseopt, reqvary, tempreqvary;  int req_caseopt, reqvary, tempreqvary;
3275  int options = *optionsptr;  int options = *optionsptr;               /* May change dynamically */
3276  int after_manual_callout = 0;  int after_manual_callout = 0;
3277  int length_prevgroup = 0;  int length_prevgroup = 0;
3278  register int c;  register int c;
# Line 2970  uschar *previous_callout = NULL; Line 3290  uschar *previous_callout = NULL;
3290  uschar *save_hwm = NULL;  uschar *save_hwm = NULL;
3291  uschar classbits[32];  uschar classbits[32];
3292    
3293    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3294    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3295    dynamically as we process the pattern. */
3296    
3297  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3298  BOOL class_utf8;  BOOL class_utf8;
3299  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8 = (options & PCRE_UTF8) != 0;
# Line 2978  uschar *class_utf8data_base; Line 3302  uschar *class_utf8data_base;
3302  uschar utf8_char[6];  uschar utf8_char[6];
3303  #else  #else
3304  BOOL utf8 = FALSE;  BOOL utf8 = FALSE;
 uschar *utf8_char = NULL;  
3305  #endif  #endif
3306    
3307  #ifdef PCRE_DEBUG  #ifdef PCRE_DEBUG
# Line 3029  for (;; ptr++) Line 3352  for (;; ptr++)
3352    int subfirstbyte;    int subfirstbyte;
3353    int terminator;    int terminator;
3354    int mclength;    int mclength;
3355      int tempbracount;
3356    uschar mcbuffer[8];    uschar mcbuffer[8];
3357    
3358    /* Get next byte in the pattern */    /* Get next byte in the pattern */
# Line 3053  for (;; ptr++) Line 3377  for (;; ptr++)
3377  #ifdef PCRE_DEBUG  #ifdef PCRE_DEBUG
3378      if (code > cd->hwm) cd->hwm = code;                 /* High water info */      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3379  #endif  #endif
3380      if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */      if (code > cd->start_workspace + cd->workspace_size -
3381            WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
3382        {        {
3383        *errorcodeptr = ERR52;        *errorcodeptr = ERR52;
3384        goto FAILED;        goto FAILED;
# Line 3076  for (;; ptr++) Line 3401  for (;; ptr++)
3401        }        }
3402    
3403      *lengthptr += (int)(code - last_code);      *lengthptr += (int)(code - last_code);
3404      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),
3405          c));
3406    
3407      /* If "previous" is set and it is not at the start of the work space, move      /* If "previous" is set and it is not at the start of the work space, move
3408      it back to there, in order to avoid filling up the work space. Otherwise,      it back to there, in order to avoid filling up the work space. Otherwise,
# Line 3102  for (;; ptr++) Line 3428  for (;; ptr++)
3428    /* In the real compile phase, just check the workspace used by the forward    /* In the real compile phase, just check the workspace used by the forward
3429    reference list. */    reference list. */
3430    
3431    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)    else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3432               WORK_SIZE_SAFETY_MARGIN)
3433      {      {
3434      *errorcodeptr = ERR52;      *errorcodeptr = ERR52;
3435      goto FAILED;      goto FAILED;
# Line 3150  for (;; ptr++) Line 3477  for (;; ptr++)
3477      previous_callout = NULL;      previous_callout = NULL;
3478      }      }
3479    
3480    /* In extended mode, skip white space and comments */    /* In extended mode, skip white space and comments. */
3481    
3482    if ((options & PCRE_EXTENDED) != 0)    if ((options & PCRE_EXTENDED) != 0)
3483      {      {
3484      if ((cd->ctypes[c] & ctype_space) != 0) continue;      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3485      if (c == CHAR_NUMBER_SIGN)      if (c == CHAR_NUMBER_SIGN)
3486        {        {
3487        ptr++;        ptr++;
3488        while (*ptr != 0)        while (*ptr != 0)
3489          {          {
3490          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3491          ptr++;          ptr++;
3492  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3493          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3494  #endif  #endif
3495          }          }
# Line 3209  for (;; ptr++) Line 3536  for (;; ptr++)
3536      the setting of any following char as a first character. */      the setting of any following char as a first character. */
3537    
3538      case CHAR_CIRCUMFLEX_ACCENT:      case CHAR_CIRCUMFLEX_ACCENT:
3539        previous = NULL;
3540      if ((options & PCRE_MULTILINE) != 0)      if ((options & PCRE_MULTILINE) != 0)
3541        {        {
3542        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3543          *code++ = OP_CIRCM;
3544        }        }
3545      previous = NULL;      else *code++ = OP_CIRC;
     *code++ = OP_CIRC;  
3546      break;      break;
3547    
3548      case CHAR_DOLLAR_SIGN:      case CHAR_DOLLAR_SIGN:
3549      previous = NULL;      previous = NULL;
3550      *code++ = OP_DOLL;      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3551      break;      break;
3552    
3553      /* There can never be a first char if '.' is first, whatever happens about      /* There can never be a first char if '.' is first, whatever happens about
# Line 3355  for (;; ptr++) Line 3683  for (;; ptr++)
3683    
3684        if (lengthptr != NULL)        if (lengthptr != NULL)
3685          {          {
3686          *lengthptr += class_utf8data - class_utf8data_base;          *lengthptr += (int)(class_utf8data - class_utf8data_base);
3687          class_utf8data = class_utf8data_base;          class_utf8data = class_utf8data_base;
3688          }          }
3689    
# Line 3491  for (;; ptr++) Line 3819  for (;; ptr++)
3819          if (*errorcodeptr != 0) goto FAILED;          if (*errorcodeptr != 0) goto FAILED;
3820    
3821          if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */          if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
3822            else if (-c == ESC_N)            /* \N is not supported in a class */
3823              {
3824              *errorcodeptr = ERR71;
3825              goto FAILED;
3826              }
3827          else if (-c == ESC_Q)            /* Handle start of quoted string */          else if (-c == ESC_Q)            /* Handle start of quoted string */
3828            {            {
3829            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
# Line 3540  for (;; ptr++) Line 3873  for (;; ptr++)
3873              continue;              continue;
3874    
3875              /* Perl 5.004 onwards omits VT from \s, but we must preserve it              /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3876              if it was previously set by something earlier in the character              if it was previously set by something earlier in the character
3877              class. */              class. */
3878    
3879              case ESC_s:              case ESC_s:
3880              classbits[0] |= cbits[cbit_space];              classbits[0] |= cbits[cbit_space];
3881              classbits[1] |= cbits[cbit_space+1] & ~0x08;              classbits[1] |= cbits[cbit_space+1] & ~0x08;
3882              for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];              for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3883              continue;              continue;
3884    
# Line 3965  for (;; ptr++) Line 4298  for (;; ptr++)
4298    
4299      In UTF-8 mode, we can optimize the negative case only if there were no      In UTF-8 mode, we can optimize the negative case only if there were no
4300      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4301      operate on single-bytes only. This is an historical hangover. Maybe one day      operate on single-bytes characters only. This is an historical hangover.
4302      we can tidy these opcodes to handle multi-byte characters.      Maybe one day we can tidy these opcodes to handle multi-byte characters.
4303    
4304      The optimization throws away the bit map. We turn the item into a      The optimization throws away the bit map. We turn the item into a
4305      1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note      1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative.
4306      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
4307      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
4308      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
4309      reqbyte, save the previous value for reinstating. */      of reqbyte, save the previous value for reinstating. */
4310    
4311  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4312      if (class_charcount == 1 && !class_utf8 &&      if (class_charcount == 1 && !class_utf8 &&
# Line 3984  for (;; ptr++) Line 4317  for (;; ptr++)
4317        {        {
4318        zeroreqbyte = reqbyte;        zeroreqbyte = reqbyte;
4319    
4320        /* The OP_NOT opcode works on one-byte characters only. */        /* The OP_NOT[I] opcodes work on one-byte characters only. */
4321    
4322        if (negate_class)        if (negate_class)
4323          {          {
4324          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4325          zerofirstbyte = firstbyte;          zerofirstbyte = firstbyte;
4326          *code++ = OP_NOT;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4327          *code++ = class_lastchar;          *code++ = class_lastchar;
4328          break;          break;
4329          }          }
# Line 4049  for (;; ptr++) Line 4382  for (;; ptr++)
4382    
4383        /* Now fill in the complete length of the item */        /* Now fill in the complete length of the item */
4384    
4385        PUT(previous, 1, code - previous);        PUT(previous, 1, (int)(code - previous));
4386        break;   /* End of class handling */        break;   /* End of class handling */
4387        }        }
4388  #endif  #endif
# Line 4118  for (;; ptr++) Line 4451  for (;; ptr++)
4451      op_type = 0;                    /* Default single-char op codes */      op_type = 0;                    /* Default single-char op codes */
4452      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
4453    
4454      /* 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
4455      for an inserted OP_ONCE for the additional '+' extension. */      insert something before it. */
4456    
4457      tempcode = previous;      tempcode = previous;
4458    
# Line 4142  for (;; ptr++) Line 4475  for (;; ptr++)
4475        }        }
4476      else repeat_type = greedy_default;      else repeat_type = greedy_default;
4477    
4478        /* If previous was a recursion call, wrap it in atomic brackets so that
4479        previous becomes the atomic group. All recursions were so wrapped in the
4480        past, but it no longer happens for non-repeated recursions. In fact, the
4481        repeated ones could be re-implemented independently so as not to need this,
4482        but for the moment we rely on the code for repeating groups. */
4483    
4484        if (*previous == OP_RECURSE)
4485          {
4486          memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4487          *previous = OP_ONCE;
4488          PUT(previous, 1, 2 + 2*LINK_SIZE);
4489          previous[2 + 2*LINK_SIZE] = OP_KET;
4490          PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4491          code += 2 + 2 * LINK_SIZE;
4492          length_prevgroup = 3 + 3*LINK_SIZE;
4493    
4494          /* When actually compiling, we need to check whether this was a forward
4495          reference, and if so, adjust the offset. */
4496    
4497          if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4498            {
4499            int offset = GET(cd->hwm, -LINK_SIZE);
4500            if (offset == previous + 1 - cd->start_code)
4501              PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4502            }
4503          }
4504    
4505        /* Now handle repetition for the different types of item. */
4506    
4507      /* If previous was a character match, abolish the item and generate a      /* If previous was a character match, abolish the item and generate a
4508      repeat item instead. If a char item has a minumum of more than one, ensure      repeat item instead. If a char item has a minumum of more than one, ensure
4509      that it is set in reqbyte - it might not be if a sequence such as x{3} is      that it is set in reqbyte - it might not be if a sequence such as x{3} is
4510      the first thing in a branch because the x will have gone into firstbyte      the first thing in a branch because the x will have gone into firstbyte
4511      instead.  */      instead.  */
4512    
4513      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (*previous == OP_CHAR || *previous == OP_CHARI)
4514        {        {
4515          op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4516    
4517        /* Deal with UTF-8 characters that take up more than one byte. It's        /* Deal with UTF-8 characters that take up more than one byte. It's
4518        easier to write this out separately than try to macrify it. Use c to        easier to write this out separately than try to macrify it. Use c to
4519        hold the length of the character in bytes, plus 0x80 to flag that it's a        hold the length of the character in bytes, plus 0x80 to flag that it's a
# Line 4160  for (;; ptr++) Line 4524  for (;; ptr++)
4524          {          {
4525          uschar *lastchar = code - 1;          uschar *lastchar = code - 1;
4526          while((*lastchar & 0xc0) == 0x80) lastchar--;          while((*lastchar & 0xc0) == 0x80) lastchar--;
4527          c = code - lastchar;            /* Length of UTF-8 character */          c = (int)(code - lastchar);     /* Length of UTF-8 character */
4528          memcpy(utf8_char, lastchar, c); /* Save the char */          memcpy(utf8_char, lastchar, c); /* Save the char */
4529          c |= 0x80;                      /* Flag c as a length */          c |= 0x80;                      /* Flag c as a length */
4530          }          }
# Line 4194  for (;; ptr++) Line 4558  for (;; ptr++)
4558      /* If previous was a single negated character ([^a] or similar), we use      /* If previous was a single negated character ([^a] or similar), we use
4559      one of the special opcodes, replacing it. The code is shared with single-      one of the special opcodes, replacing it. The code is shared with single-
4560      character repeats by setting opt_type to add a suitable offset into      character repeats by setting opt_type to add a suitable offset into
4561      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
4562      currently used only for single-byte chars. */      are currently used only for single-byte chars. */
4563    
4564      else if (*previous == OP_NOT)      else if (*previous == OP_NOT || *previous == OP_NOTI)
4565        {        {
4566        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4567        c = previous[1];        c = previous[1];
4568        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4569            repeat_max < 0 &&            repeat_max < 0 &&
# Line 4396  for (;; ptr++) Line 4760  for (;; ptr++)
4760  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4761               *previous == OP_XCLASS ||               *previous == OP_XCLASS ||
4762  #endif  #endif
4763               *previous == OP_REF)               *previous == OP_REF ||
4764                 *previous == OP_REFI)
4765        {        {
4766        if (repeat_max == 0)        if (repeat_max == 0)
4767          {          {
# Line 4430  for (;; ptr++) Line 4795  for (;; ptr++)
4795        }        }
4796    
4797      /* If previous was a bracket group, we may have to replicate it in certain      /* If previous was a bracket group, we may have to replicate it in certain
4798      cases. */      cases. Note that at this point we can encounter only the "basic" bracket
4799        opcodes such as BRA and CBRA, as this is the place where they get converted
4800        into the more special varieties such as BRAPOS and SBRA. A test for >=
4801        OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4802        ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4803        repetition of assertions, but now it does, for Perl compatibility. */
4804    
4805      else if (*previous == OP_BRA  || *previous == OP_CBRA ||      else if (*previous >= OP_ASSERT && *previous <= OP_COND)
              *previous == OP_ONCE || *previous == OP_COND)  
4806        {        {
4807        register int i;        register int i;
       int ketoffset = 0;  
4808        int len = (int)(code - previous);        int len = (int)(code - previous);
4809        uschar *bralink = NULL;        uschar *bralink = NULL;
4810          uschar *brazeroptr = NULL;
4811    
4812        /* Repeating a DEFINE group is pointless */        /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4813          we just ignore the repeat. */
4814    
4815        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4816          {          goto END_REPEAT;
4817          *errorcodeptr = ERR55;  
4818          goto FAILED;        /* There is no sense in actually repeating assertions. The only potential
4819          }        use of repetition is in cases when the assertion is optional. Therefore,
4820          if the minimum is greater than zero, just ignore the repeat. If the
4821          maximum is not not zero or one, set it to 1. */
4822    
4823        /* If the maximum repeat count is unlimited, find the end of the bracket        if (*previous < OP_ONCE)    /* Assertion */
4824        by scanning through from the start, and compute the offset back to it          {
4825        from the current code pointer. There may be an OP_OPT setting following          if (repeat_min > 0) goto END_REPEAT;
4826        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);  
4827          }          }
4828    
4829        /* The case of a zero minimum is special because of the need to stick        /* The case of a zero minimum is special because of the need to stick
# Line 4479  for (;; ptr++) Line 4844  for (;; ptr++)
4844          **   goto END_REPEAT;          **   goto END_REPEAT;
4845          **   }          **   }
4846    
4847          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
4848          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
4849          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
4850          groups are referenced, we cannot do this selectively.          don't have a list of which groups are referenced, we cannot do this
4851            selectively.
4852    
4853          If the maximum is 1 or unlimited, we just have to stick in the BRAZERO          If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4854          and do no more at this point. However, we do need to adjust any          and do no more at this point. However, we do need to adjust any
# Line 4502  for (;; ptr++) Line 4868  for (;; ptr++)
4868              *previous++ = OP_SKIPZERO;              *previous++ = OP_SKIPZERO;
4869              goto END_REPEAT;              goto END_REPEAT;
4870              }              }
4871              brazeroptr = previous;    /* Save for possessive optimizing */
4872            *previous++ = OP_BRAZERO + repeat_type;            *previous++ = OP_BRAZERO + repeat_type;
4873            }            }
4874    
# Line 4564  for (;; ptr++) Line 4931  for (;; ptr++)
4931              *lengthptr += delta;              *lengthptr += delta;
4932              }              }
4933    
4934            /* This is compiling for real */            /* This is compiling for real. If there is a set first byte for
4935              the group, and we have not yet set a "required byte", set it. Make
4936              sure there is enough workspace for copying forward references before
4937              doing the copy. */
4938    
4939            else            else
4940              {              {
4941              if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;              if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4942    
4943              for (i = 1; i < repeat_min; i++)              for (i = 1; i < repeat_min; i++)
4944                {                {
4945                uschar *hc;                uschar *hc;
4946                uschar *this_hwm = cd->hwm;                uschar *this_hwm = cd->hwm;
4947                memcpy(code, previous, len);                memcpy(code, previous, len);
4948    
4949                  while (cd->hwm > cd->start_workspace + cd->workspace_size -
4950                         WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
4951                    {
4952                    int save_offset = save_hwm - cd->start_workspace;
4953                    int this_offset = this_hwm - cd->start_workspace;
4954                    *errorcodeptr = expand_workspace(cd);
4955                    if (*errorcodeptr != 0) goto FAILED;
4956                    save_hwm = (uschar *)cd->start_workspace + save_offset;
4957                    this_hwm = (uschar *)cd->start_workspace + this_offset;
4958                    }
4959    
4960                for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)                for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4961                  {                  {
4962                  PUT(cd->hwm, 0, GET(hc, 0) + len);                  PUT(cd->hwm, 0, GET(hc, 0) + len);
# Line 4603  for (;; ptr++) Line 4986  for (;; ptr++)
4986          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4987          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4988          a 64-bit integer type when available, otherwise double. */          a 64-bit integer type when available, otherwise double. */
4989    
4990          if (lengthptr != NULL && repeat_max > 0)          if (lengthptr != NULL && repeat_max > 0)
4991            {            {
4992            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
# Line 4641  for (;; ptr++) Line 5024  for (;; ptr++)
5024              }              }
5025    
5026            memcpy(code, previous, len);            memcpy(code, previous, len);
5027    
5028              /* Ensure there is enough workspace for forward references before
5029              copying them. */
5030    
5031              while (cd->hwm > cd->start_workspace + cd->workspace_size -
5032                     WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5033                {
5034                int save_offset = save_hwm - cd->start_workspace;
5035                int this_offset = this_hwm - cd->start_workspace;
5036                *errorcodeptr = expand_workspace(cd);
5037                if (*errorcodeptr != 0) goto FAILED;
5038                save_hwm = (uschar *)cd->start_workspace + save_offset;
5039                this_hwm = (uschar *)cd->start_workspace + this_offset;
5040                }
5041    
5042            for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)            for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5043              {              {
5044              PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));              PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
# Line 4666  for (;; ptr++) Line 5064  for (;; ptr++)
5064            }            }
5065          }          }
5066    
5067        /* 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
5068        can't just offset backwards from the current code point, because we        ONCE brackets, that's all we need to do. However, possessively repeated
5069        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
5070        correct offset was computed above.        behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5071          deal with possessive ONCEs specially.
5072        Then, when we are doing the actual compile phase, check to see whether  
5073        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
5074          whether this group is one that could match an empty string. If so,
5075        convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so        convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
5076        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
5077        atomic groups at runtime, but in a different way.] */        groups at runtime, but in a different way.]
5078    
5079          Then, if the quantifier was possessive and the bracket is not a
5080          conditional, we convert the BRA code to the POS form, and the KET code to
5081          KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5082          subpattern at both the start and at the end.) The use of special opcodes
5083          makes it possible to reduce greatly the stack usage in pcre_exec(). If
5084          the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5085    
5086          Then, if the minimum number of matches is 1 or 0, cancel the possessive
5087          flag so that the default action below, of wrapping everything inside
5088          atomic brackets, does not happen. When the minimum is greater than 1,
5089          there will be earlier copies of the group, and so we still have to wrap
5090          the whole thing. */
5091    
5092        else        else
5093          {          {
5094          uschar *ketcode = code - ketoffset;          uschar *ketcode = code - 1 - LINK_SIZE;
5095          uschar *bracode = ketcode - GET(ketcode, 1);          uschar *bracode = ketcode - GET(ketcode, 1);
5096          *ketcode = OP_KETRMAX + repeat_type;  
5097          if (lengthptr == NULL && *bracode != OP_ONCE)          /* Convert possessive ONCE brackets to non-capturing */
5098    
5099            if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5100                possessive_quantifier) *bracode = OP_BRA;
5101    
5102            /* For non-possessive ONCE brackets, all we need to do is to
5103            set the KET. */
5104    
5105            if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5106              *ketcode = OP_KETRMAX + repeat_type;
5107    
5108            /* Handle non-ONCE brackets and possessive ONCEs (which have been
5109            converted to non-capturing above). */
5110    
5111            else
5112            {            {
5113            uschar *scode = bracode;            /* In the compile phase, check for empty string matching. */
5114            do  
5115              if (lengthptr == NULL)
5116              {              {
5117              if (could_be_empty_branch(scode, ketcode, utf8, cd))              uschar *scode = bracode;
5118                do
5119                {                {
5120                *bracode += OP_SBRA - OP_BRA;                if (could_be_empty_branch(scode, ketcode, utf8, cd))
5121                break;                  {
5122                    *bracode += OP_SBRA - OP_BRA;
5123                    break;
5124                    }
5125                  scode += GET(scode, 1);
5126                }                }
5127              scode += GET(scode, 1);              while (*scode == OP_ALT);
5128              }              }
5129            while (*scode == OP_ALT);  
5130              /* Handle possessive quantifiers. */
5131    
5132              if (possessive_quantifier)
5133                {
5134                /* For COND brackets, we wrap the whole thing in a possessively
5135                repeated non-capturing bracket, because we have not invented POS
5136                versions of the COND opcodes. Because we are moving code along, we
5137                must ensure that any pending recursive references are updated. */
5138    
5139                if (*bracode == OP_COND || *bracode == OP_SCOND)
5140                  {
5141                  int nlen = (int)(code - bracode);
5142                  *code = OP_END;
5143                  adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5144                  memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5145                  code += 1 + LINK_SIZE;
5146                  nlen += 1 + LINK_SIZE;
5147                  *bracode = OP_BRAPOS;
5148                  *code++ = OP_KETRPOS;
5149                  PUTINC(code, 0, nlen);
5150                  PUT(bracode, 1, nlen);
5151                  }
5152    
5153                /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5154    
5155                else
5156                  {
5157                  *bracode += 1;              /* Switch to xxxPOS opcodes */
5158                  *ketcode = OP_KETRPOS;
5159                  }
5160    
5161                /* If the minimum is zero, mark it as possessive, then unset the
5162                possessive flag when the minimum is 0 or 1. */
5163    
5164                if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5165                if (repeat_min < 2) possessive_quantifier = FALSE;
5166                }
5167    
5168              /* Non-possessive quantifier */
5169    
5170              else *ketcode = OP_KETRMAX + repeat_type;
5171            }            }
5172          }          }
5173        }        }
# Line 4715  for (;; ptr++) Line 5188  for (;; ptr++)
5188        }        }
5189    
5190      /* If the character following a repeat is '+', or if certain optimization      /* If the character following a repeat is '+', or if certain optimization
5191      tests above succeeded, possessive_quantifier is TRUE. For some of the      tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5192      simpler opcodes, there is an special alternative opcode for this. For      there are special alternative opcodes for this case. For anything else, we
5193      anything else, we wrap the entire repeated item inside OP_ONCE brackets.      wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5194      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
5195      but the special opcodes can optimize it a bit. The repeated item starts at      special opcodes can optimize it.
5196      tempcode, not at previous, which might be the first part of a string whose  
5197      (former) last char we repeated.      Some (but not all) possessively repeated subpatterns have already been
5198        completely handled in the code just above. For them, possessive_quantifier
5199        is always FALSE at this stage.
5200    
5201        Note that the repeated item starts at tempcode, not at previous, which
5202        might be the first part of a string whose (former) last char we repeated.
5203    
5204      Possessifying an 'exact' quantifier has no effect, so we can ignore it. But      Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
5205      an 'upto' may follow. We skip over an 'exact' item, and then test the      an 'upto' may follow. We skip over an 'exact' item, and then test the
# Line 4752  for (;; ptr++) Line 5230  for (;; ptr++)
5230          case OP_QUERY: *tempcode = OP_POSQUERY; break;          case OP_QUERY: *tempcode = OP_POSQUERY; break;
5231          case OP_UPTO:  *tempcode = OP_POSUPTO; break;          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
5232    
5233          case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;          case OP_STARI:  *tempcode = OP_POSSTARI; break;
5234          case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;          case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
5235          case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;          case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5236          case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;          case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
5237    
5238          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
5239          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
5240          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5241          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
5242    
5243            case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
5244            case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
5245            case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5246            case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
5247    
5248            case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
5249            case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
5250            case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5251            case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
5252    
5253          /* Because we are moving code along, we must ensure that any          /* Because we are moving code along, we must ensure that any
5254          pending recursive references are updated. */          pending recursive references are updated. */
5255    
# Line 4815  for (;; ptr++) Line 5303  for (;; ptr++)
5303        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5304        namelen = (int)(ptr - name);        namelen = (int)(ptr - name);
5305    
5306          /* It appears that Perl allows any characters whatsoever, other than
5307          a closing parenthesis, to appear in arguments, so we no longer insist on
5308          letters, digits, and underscores. */
5309    
5310        if (*ptr == CHAR_COLON)        if (*ptr == CHAR_COLON)
5311          {          {
5312          arg = ++ptr;          arg = ++ptr;
5313          while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
           || *ptr == '_') ptr++;  
5314          arglen = (int)(ptr - arg);          arglen = (int)(ptr - arg);
5315          }          }
5316    
# Line 4836  for (;; ptr++) Line 5327  for (;; ptr++)
5327          if (namelen == verbs[i].len &&          if (namelen == verbs[i].len &&
5328              strncmp((char *)name, vn, namelen) == 0)              strncmp((char *)name, vn, namelen) == 0)
5329            {            {
5330            /* Check for open captures before ACCEPT */            /* Check for open captures before ACCEPT and convert it to
5331              ASSERT_ACCEPT if in an assertion. */
5332    
5333            if (verbs[i].op == OP_ACCEPT)            if (verbs[i].op == OP_ACCEPT)
5334              {              {
5335              open_capitem *oc;              open_capitem *oc;
5336                if (arglen != 0)
5337                  {
5338                  *errorcodeptr = ERR59;
5339                  goto FAILED;
5340                  }
5341              cd->had_accept = TRUE;              cd->had_accept = TRUE;
5342              for (oc = cd->open_caps; oc != NULL; oc = oc->next)              for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5343                {                {
5344                *code++ = OP_CLOSE;                *code++ = OP_CLOSE;
5345                PUT2INC(code, 0, oc->number);                PUT2INC(code, 0, oc->number);
5346                }                }
5347                *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5348    
5349                /* Do not set firstbyte after *ACCEPT */
5350                if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5351              }              }
5352    
5353            /* Handle the cases with/without an argument */            /* Handle other cases with/without an argument */
5354    
5355            if (arglen == 0)            else if (arglen == 0)
5356              {              {
5357              if (verbs[i].op < 0)   /* Argument is mandatory */              if (verbs[i].op < 0)   /* Argument is mandatory */
5358                {                {
# Line 4859  for (;; ptr++) Line 5360  for (;; ptr++)
5360                goto FAILED;                goto FAILED;
5361                }                }
5362              *code = verbs[i].op;              *code = verbs[i].op;
5363              if (*code++ == OP_THEN)              if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
               {  
               PUT(code, 0, code - bcptr->current_branch - 1);  
               code += LINK_SIZE;  
               }  
5364              }              }
5365    
5366            else            else
# Line 4874  for (;; ptr++) Line 5371  for (;; ptr++)
5371                goto FAILED;                goto FAILED;
5372                }                }
5373              *code = verbs[i].op_arg;              *code = verbs[i].op_arg;
5374              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;  
               }  
5375              *code++ = arglen;              *code++ = arglen;
5376              memcpy(code, arg, arglen);              memcpy(code, arg, arglen);
5377              code += arglen;              code += arglen;
# Line 5140  for (;; ptr++) Line 5633  for (;; ptr++)
5633          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5634          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
5635          bravalue = OP_ASSERT;          bravalue = OP_ASSERT;
5636            cd->assert_depth += 1;
5637          ptr++;          ptr++;
5638          break;          break;
5639    
# Line 5154  for (;; ptr++) Line 5648  for (;; ptr++)
5648            continue;            continue;
5649            }            }
5650          bravalue = OP_ASSERT_NOT;          bravalue = OP_ASSERT_NOT;
5651            cd->assert_depth += 1;
5652          break;          break;
5653    
5654    
# Line 5163  for (;; ptr++) Line 5658  for (;; ptr++)
5658            {            {
5659            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
5660            bravalue = OP_ASSERTBACK;            bravalue = OP_ASSERTBACK;
5661              cd->assert_depth += 1;
5662            ptr += 2;            ptr += 2;
5663            break;            break;
5664    
5665            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
5666            bravalue = OP_ASSERTBACK_NOT;            bravalue = OP_ASSERTBACK_NOT;
5667              cd->assert_depth += 1;
5668            ptr += 2;            ptr += 2;
5669            break;            break;
5670    
# Line 5189  for (;; ptr++) Line 5686  for (;; ptr++)
5686    
5687          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5688          case CHAR_C:                 /* Callout - may be followed by digits; */          case CHAR_C:                 /* Callout - may be followed by digits; */
5689          previous_callout = code;  /* Save for later completion */          previous_callout = code;     /* Save for later completion */
5690          after_manual_callout = 1; /* Skip one item before completing */          after_manual_callout = 1;    /* Skip one item before completing */
5691          *code++ = OP_CALLOUT;          *code++ = OP_CALLOUT;
5692            {            {
5693            int n = 0;            int n = 0;
# Line 5373  for (;; ptr++) Line 5870  for (;; ptr++)
5870          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5871          namelen = (int)(ptr - name);          namelen = (int)(ptr - name);
5872    
5873          /* 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
5874          reference number. */          a dummy reference number, because it was not used in the first pass.
5875            However, with the change of recursive back references to be atomic,
5876            we have to look for the number so that this state can be identified, as
5877            otherwise the incorrect length is computed. If it's not a backwards
5878            reference, the dummy number will do. */
5879    
5880          if (lengthptr != NULL)          if (lengthptr != NULL)
5881            {            {
5882              const uschar *temp;
5883    
5884            if (namelen == 0)            if (namelen == 0)
5885              {              {
5886              *errorcodeptr = ERR62;              *errorcodeptr = ERR62;
# Line 5393  for (;; ptr++) Line 5896  for (;; ptr++)
5896              *errorcodeptr = ERR48;              *errorcodeptr = ERR48;
5897              goto FAILED;              goto FAILED;
5898              }              }
5899            recno = 0;  
5900              /* The name table does not exist in the first pass, so we cannot
5901              do a simple search as in the code below. Instead, we have to scan the
5902              pattern to find the number. It is important that we scan it only as
5903              far as we have got because the syntax of named subpatterns has not
5904              been checked for the rest of the pattern, and find_parens() assumes
5905              correct syntax. In any case, it's a waste of resources to scan
5906              further. We stop the scan at the current point by temporarily
5907              adjusting the value of cd->endpattern. */
5908    
5909              temp = cd->end_pattern;
5910              cd->end_pattern = ptr;
5911              recno = find_parens(cd, name, namelen,
5912                (options & PCRE_EXTENDED) != 0, utf8);
5913              cd->end_pattern = temp;
5914              if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
5915            }            }
5916    
5917          /* In the real compile, seek the name in the table. We check the name          /* In the real compile, seek the name in the table. We check the name
# Line 5537  for (;; ptr++) Line 6055  for (;; ptr++)
6055    
6056                /* Fudge the value of "called" so that when it is inserted as an                /* Fudge the value of "called" so that when it is inserted as an
6057                offset below, what it actually inserted is the reference number                offset below, what it actually inserted is the reference number
6058                of the group. */                of the group. Then remember the forward reference. */
6059    
6060                called = cd->start_code + recno;                called = cd->start_code + recno;
6061                PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));                if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6062                      WORK_SIZE_SAFETY_MARGIN)
6063                    {
6064                    *errorcodeptr = expand_workspace(cd);
6065                    if (*errorcodeptr != 0) goto FAILED;
6066                    }
6067                  PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6068                }                }
6069    
6070              /* If not a forward reference, and the subpattern is still open,              /* If not a forward reference, and the subpattern is still open,
6071              this is a recursive call. We check to see if this is a left              this is a recursive call. We check to see if this is a left
6072              recursion that could loop for ever, and diagnose that case. */              recursion that could loop for ever, and diagnose that case. We
6073                must not, however, do this check if we are in a conditional
6074                subpattern because the condition might be testing for recursion in
6075                a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6076                Forever loops are also detected at runtime, so those that occur in
6077                conditional subpatterns will be picked up then. */
6078    
6079              else if (GET(called, 1) == 0 &&              else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6080                       could_be_empty(called, code, bcptr, utf8, cd))                       could_be_empty(called, code, bcptr, utf8, cd))
6081                {                {
6082                *errorcodeptr = ERR40;                *errorcodeptr = ERR40;
# Line 5555  for (;; ptr++) Line 6084  for (;; ptr++)
6084                }                }
6085              }              }
6086    
6087            /* Insert the recursion/subroutine item, automatically wrapped inside            /* Insert the recursion/subroutine item. It does not have a set first
6088            "once" brackets. Set up a "previous group" length so that a            byte (relevant if it is repeated, because it will then be wrapped
6089            subsequent quantifier will work. */            with ONCE brackets). */
   
           *code = OP_ONCE;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
6090    
6091            *code = OP_RECURSE;            *code = OP_RECURSE;
6092            PUT(code, 1, (int)(called - cd->start_code));            PUT(code, 1, (int)(called - cd->start_code));
6093            code += 1 + LINK_SIZE;            code += 1 + LINK_SIZE;
6094              groupsetfirstbyte = FALSE;
           *code = OP_KET;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
   
           length_prevgroup = 3 + 3*LINK_SIZE;  
6095            }            }
6096    
6097          /* Can't determine a first byte now */          /* Can't determine a first byte now */
# Line 5632  for (;; ptr++) Line 6152  for (;; ptr++)
6152          is necessary to ensure we correctly detect the start of the pattern in          is necessary to ensure we correctly detect the start of the pattern in
6153          both phases.          both phases.
6154    
6155          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
6156          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. */  
6157    
6158          if (*ptr == CHAR_RIGHT_PARENTHESIS)          if (*ptr == CHAR_RIGHT_PARENTHESIS)
6159            {            {
# Line 5645  for (;; ptr++) Line 6164  for (;; ptr++)
6164              }              }
6165            else            else
6166              {              {
             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))  
               {  
               *code++ = OP_OPT;  
               *code++ = newoptions & PCRE_IMS;  
               }  
6167              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6168              greedy_non_default = greedy_default ^ 1;              greedy_non_default = greedy_default ^ 1;
6169              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6170              }              }
6171    
6172            /* Change options at this level, and pass them back for use            /* Change options at this level, and pass them back for use
6173            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). */  
6174    
6175            *optionsptr = options = newoptions;            *optionsptr = options = newoptions;
6176            previous = NULL;       /* This item can't be repeated */            previous = NULL;       /* This item can't be repeated */
# Line 5694  for (;; ptr++) Line 6206  for (;; ptr++)
6206        skipbytes = 2;        skipbytes = 2;
6207        }        }
6208    
6209      /* Process nested bracketed regex. Assertions may not be repeated, but      /* Process nested bracketed regex. Assertions used not to be repeatable,
6210      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
6211      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
6212      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. */  
6213    
6214      previous = (bravalue >= OP_ONCE)? code : NULL;      previous = code;                      /* For handling repetition */
6215      *code = bravalue;      *code = bravalue;
6216      tempcode = code;      tempcode = code;
6217      tempreqvary = cd->req_varyopt;     /* Save value before bracket */      tempreqvary = cd->req_varyopt;        /* Save value before bracket */
6218      length_prevgroup = 0;              /* Initialize for pre-compile phase */      tempbracount = cd->bracount;          /* Save value before bracket */
6219        length_prevgroup = 0;                 /* Initialize for pre-compile phase */
6220    
6221      if (!compile_regex(      if (!compile_regex(
6222           newoptions,                   /* The complete new option state */           newoptions,                      /* The complete new option state */
6223           options & PCRE_IMS,           /* The previous ims option state */           &tempcode,                       /* Where to put code (updated) */
6224           &tempcode,                    /* Where to put code (updated) */           &ptr,                            /* Input pointer (updated) */
6225           &ptr,                         /* Input pointer (updated) */           errorcodeptr,                    /* Where to put an error message */
          errorcodeptr,                 /* Where to put an error message */  
6226           (bravalue == OP_ASSERTBACK ||           (bravalue == OP_ASSERTBACK ||
6227            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6228           reset_bracount,               /* True if (?| group */           reset_bracount,                  /* True if (?| group */
6229           skipbytes,                    /* Skip over bracket number */           skipbytes,                       /* Skip over bracket number */
6230           &subfirstbyte,                /* For possible first char */           cond_depth +
6231           &subreqbyte,                  /* For possible last char */             ((bravalue == OP_COND)?1:0),   /* Depth of condition subpatterns */
6232           bcptr,                        /* Current branch chain */           &subfirstbyte,                   /* For possible first char */
6233           cd,                           /* Tables block */           &subreqbyte,                     /* For possible last char */
6234           (lengthptr == NULL)? NULL :   /* Actual compile phase */           bcptr,                           /* Current branch chain */
6235             &length_prevgroup           /* Pre-compile phase */           cd,                              /* Tables block */
6236             (lengthptr == NULL)? NULL :      /* Actual compile phase */
6237               &length_prevgroup              /* Pre-compile phase */
6238           ))           ))
6239        goto FAILED;        goto FAILED;
6240    
6241        /* If this was an atomic group and there are no capturing groups within it,
6242        generate OP_ONCE_NC instead of OP_ONCE. */
6243    
6244        if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6245          *code = OP_ONCE_NC;
6246    
6247        if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6248          cd->assert_depth -= 1;
6249    
6250      /* At the end of compiling, code is still pointing to the start of the      /* At the end of compiling, code is still pointing to the start of the
6251      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.
6252      and any option resetting that may follow it. The pattern pointer (ptr)      The pattern pointer (ptr) is on the bracket.
     is on the bracket. */  
6253    
6254      /* 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
6255      two branches in the group, or just one if it's a DEFINE group. We do this      two branches in the group, or just one if it's a DEFINE group. We do this
6256      in the real compile phase, not in the pre-pass, where the whole group may      in the real compile phase, not in the pre-pass, where the whole group may
6257      not be available. */      not be available. */
# Line 5795  for (;; ptr++) Line 6316  for (;; ptr++)
6316          goto FAILED;          goto FAILED;
6317          }          }
6318        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6319        *code++ = OP_BRA;        code++;   /* This already contains bravalue */
6320        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
6321        *code++ = OP_KET;        *code++ = OP_KET;
6322        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
# Line 5963  for (;; ptr++) Line 6484  for (;; ptr++)
6484          }          }
6485    
6486        /* \k<name> or \k'name' is a back reference by name (Perl syntax).        /* \k<name> or \k'name' is a back reference by name (Perl syntax).
6487        We also support \k{name} (.NET syntax) */        We also support \k{name} (.NET syntax).  */
6488    
6489        if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||        if (-c == ESC_k)
           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))  
6490          {          {
6491            if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6492              ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6493              {
6494              *errorcodeptr = ERR69;
6495              break;
6496              }
6497          is_recurse = FALSE;          is_recurse = FALSE;
6498          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6499            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
# Line 5987  for (;; ptr++) Line 6513  for (;; ptr++)
6513          HANDLE_REFERENCE:    /* Come here from named backref handling */          HANDLE_REFERENCE:    /* Come here from named backref handling */
6514          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6515          previous = code;          previous = code;
6516          *code++ = OP_REF;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6517          PUT2INC(code, 0, recno);          PUT2INC(code, 0, recno);
6518          cd->backref_map |= (recno < 32)? (1 << recno) : 1;          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6519          if (recno > cd->top_backref) cd->top_backref = recno;          if (recno > cd->top_backref) cd->top_backref = recno;
# Line 6047  for (;; ptr++) Line 6573  for (;; ptr++)
6573            }            }
6574          else          else
6575  #endif  #endif
6576            {          /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6577            so that it works in DFA mode and in lookbehinds. */
6578    
6579              {
6580            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6581            *code++ = -c;            *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6582            }            }
6583          }          }
6584        continue;        continue;
# Line 6095  for (;; ptr++) Line 6624  for (;; ptr++)
6624    
6625      ONE_CHAR:      ONE_CHAR:
6626      previous = code;      previous = code;
6627      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6628      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6629    
6630      /* Remember if \r or \n were seen */      /* Remember if \r or \n were seen */
# Line 6124  for (;; ptr++) Line 6653  for (;; ptr++)
6653        else firstbyte = reqbyte = REQ_NONE;        else firstbyte = reqbyte = REQ_NONE;
6654        }        }
6655    
6656      /* 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
6657      1 or the matching is caseful. */      1 or the matching is caseful. */
6658    
6659      else      else
# Line 6159  return FALSE; Line 6688  return FALSE;
6688  /* On entry, ptr is pointing past the bracket character, but on return it  /* On entry, ptr is pointing past the bracket character, but on return it
6689  points to the closing bracket, or vertical bar, or end of string. The code  points to the closing bracket, or vertical bar, or end of string. The code
6690  variable is pointing at the byte into which the BRA operator has been stored.  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.  
   
6691  This function is used during the pre-compile phase when we are trying to find  This function is used during the pre-compile phase when we are trying to find
6692  out the amount of memory needed, as well as during the real compile phase. The  out the amount of memory needed, as well as during the real compile phase. The
6693  value of lengthptr distinguishes the two phases.  value of lengthptr distinguishes the two phases.
6694    
6695  Arguments:  Arguments:
6696    options        option bits, including any changes for this subpattern    options        option bits, including any changes for this subpattern
   oldims         previous settings of ims option bits  
6697    codeptr        -> the address of the current code pointer    codeptr        -> the address of the current code pointer
6698    ptrptr         -> the address of the current pattern pointer    ptrptr         -> the address of the current pattern pointer
6699    errorcodeptr   -> pointer to error code variable    errorcodeptr   -> pointer to error code variable
6700    lookbehind     TRUE if this is a lookbehind assertion    lookbehind     TRUE if this is a lookbehind assertion
6701    reset_bracount TRUE to reset the count for each branch    reset_bracount TRUE to reset the count for each branch
6702    skipbytes      skip this many bytes at start (for brackets and OP_COND)    skipbytes      skip this many bytes at start (for brackets and OP_COND)
6703      cond_depth     depth of nesting for conditional subpatterns
6704    firstbyteptr   place to put the first required character, or a negative number    firstbyteptr   place to put the first required character, or a negative number
6705    reqbyteptr     place to put the last required character, or a negative number    reqbyteptr     place to put the last required character, or a negative number
6706    bcptr          pointer to the chain of currently open branches    bcptr          pointer to the chain of currently open branches
# Line 6188  Returns:         TRUE on success Line 6712  Returns:         TRUE on success
6712  */  */
6713    
6714  static BOOL  static BOOL
6715  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,  compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6716    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6717    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,    int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6718    int *lengthptr)    compile_data *cd, int *lengthptr)
6719  {  {
6720  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
6721  uschar *code = *codeptr;  uschar *code = *codeptr;
# Line 6205  int branchfirstbyte, branchreqbyte; Line 6729  int branchfirstbyte, branchreqbyte;
6729  int length;  int length;
6730  int orig_bracount;  int orig_bracount;
6731  int max_bracount;  int max_bracount;
 int old_external_options = cd->external_options;  
6732  branch_chain bc;  branch_chain bc;
6733    
6734  bc.outer = bcptr;  bc.outer = bcptr;
# Line 6229  pre-compile phase to find out whether an Line 6752  pre-compile phase to find out whether an
6752    
6753  /* If this is a capturing subpattern, add to the chain of open capturing items  /* If this is a capturing subpattern, add to the chain of open capturing items
6754  so that we can detect them if (*ACCEPT) is encountered. This is also used to  so that we can detect them if (*ACCEPT) is encountered. This is also used to
6755  detect groups that contain recursive back references to themselves. */  detect groups that contain recursive back references to themselves. Note that
6756    only OP_CBRA need be tested here; changing this opcode to one of its variants,
6757    e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6758    
6759  if (*code == OP_CBRA)  if (*code == OP_CBRA)
6760    {    {
# Line 6255  for (;;) Line 6780  for (;;)
6780    
6781    if (reset_bracount) cd->bracount = orig_bracount;    if (reset_bracount) cd->bracount = orig_bracount;
6782    
   /* 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;  
     }  
   
6783    /* Set up dummy OP_REVERSE if lookbehind assertion */    /* Set up dummy OP_REVERSE if lookbehind assertion */
6784    
6785    if (lookbehind)    if (lookbehind)
# Line 6278  for (;;) Line 6794  for (;;)
6794    into the length. */    into the length. */
6795    
6796    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6797          &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))          &branchreqbyte, &bc, cond_depth, cd,
6798            (lengthptr == NULL)? NULL : &length))
6799      {      {
6800      *ptrptr = ptr;      *ptrptr = ptr;
6801      return FALSE;      return FALSE;
6802      }      }
6803    
   /* 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;  
   
6804    /* Keep the highest bracket count in case (?| was used and some branch    /* Keep the highest bracket count in case (?| was used and some branch
6805    has fewer than the rest. */    has fewer than the rest. */
6806    
# Line 6353  for (;;) Line 6861  for (;;)
6861        {        {
6862        int fixed_length;        int fixed_length;
6863        *code = OP_END;        *code = OP_END;
6864        fixed_length = find_fixedlength(last_branch, options, FALSE, cd);        fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
6865            FALSE, cd);
6866        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
6867        if (fixed_length == -3)        if (fixed_length == -3)
6868          {          {
# Line 6361  for (;;) Line 6870  for (;;)
6870          }          }
6871        else if (fixed_length < 0)        else if (fixed_length < 0)
6872          {          {
6873          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;          *errorcodeptr = (fixed_length == -2)? ERR36 :
6874                            (fixed_length == -4)? ERR70: ERR25;
6875          *ptrptr = ptr;          *ptrptr = ptr;
6876          return FALSE;          return FALSE;
6877          }          }
# Line 6374  for (;;) Line 6884  for (;;)
6884    of offsets, with the field in the BRA item now becoming an offset to the    of offsets, with the field in the BRA item now becoming an offset to the
6885    first alternative. If there are no alternatives, it points to the end of the    first alternative. If there are no alternatives, it points to the end of the
6886    group. The length in the terminating ket is always the length of the whole    group. The length in the terminating ket is always the length of the whole
6887    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. */  
6888    
6889    if (*ptr != CHAR_VERTICAL_LINE)    if (*ptr != CHAR_VERTICAL_LINE)
6890      {      {
# Line 6420  for (;;) Line 6928  for (;;)
6928        cd->open_caps = cd->open_caps->next;        cd->open_caps = cd->open_caps->next;
6929        }        }
6930    
     /* Reset options if needed. */  
   
     if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)  
       {  
       *code++ = OP_OPT;  
       *code++ = oldims;  
       length += 2;  
       }  
   
6931      /* Retain the highest bracket number, in case resetting was used. */      /* Retain the highest bracket number, in case resetting was used. */
6932    
6933      cd->bracount = max_bracount;      cd->bracount = max_bracount;
# Line 6488  for (;;) Line 6987  for (;;)
6987  /* Try to find out if this is an anchored regular expression. Consider each  /* Try to find out if this is an anchored regular expression. Consider each
6988  alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket  alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6989  all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then  all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6990  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
6991  counts, since OP_CIRC can match in the middle.  be found, because ^ generates OP_CIRCM in that mode.
6992    
6993  We can also consider a regex to be anchored if OP_SOM starts all its branches.  We can also consider a regex to be anchored if OP_SOM starts all its branches.
6994  This is the code for \G, which means "match at start of match position, taking  This is the code for \G, which means "match at start of match position, taking
# Line 6510  of the more common cases more precisely. Line 7009  of the more common cases more precisely.
7009    
7010  Arguments:  Arguments:
7011    code           points to start of expression (the bracket)    code           points to start of expression (the bracket)
   options        points to the options setting  
7012    bracket_map    a bitmap of which brackets we are inside while testing; this    bracket_map    a bitmap of which brackets we are inside while testing; this
7013                    handles up to substring 31; after that we just have to take                    handles up to substring 31; after that we just have to take
7014                    the less precise approach                    the less precise approach
# Line 6520  Returns:     TRUE or FALSE Line 7018  Returns:     TRUE or FALSE
7018  */  */
7019    
7020  static BOOL  static BOOL
7021  is_anchored(register const uschar *code, int *options, unsigned int bracket_map,  is_anchored(register const uschar *code, unsigned int bracket_map,
7022    unsigned int backref_map)    unsigned int backref_map)
7023  {  {
7024  do {  do {
7025     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
7026       options, PCRE_MULTILINE, FALSE);       FALSE);
7027     register int op = *scode;     register int op = *scode;
7028    
7029     /* Non-capturing brackets */     /* Non-capturing brackets */
7030    
7031     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
7032           op == OP_SBRA || op == OP_SBRAPOS)
7033       {       {
7034       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
7035       }       }
7036    
7037     /* Capturing brackets */     /* Capturing brackets */
7038    
7039     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
7040                op == OP_SCBRA || op == OP_SCBRAPOS)
7041       {       {
7042       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
7043       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
7044       if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;       if (!is_anchored(scode, new_map, backref_map)) return FALSE;
7045       }       }
7046    
7047     /* Other brackets */     /* Other brackets */
7048    
7049     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC ||
7050                op == OP_COND)
7051       {       {
7052       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
7053       }       }
7054    
7055     /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and     /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
# Line 6563  do { Line 7064  do {
7064    
7065     /* Check for explicit anchoring */     /* Check for explicit anchoring */
7066    
7067     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;  
7068     code += GET(code, 1);     code += GET(code, 1);
7069     }     }
7070  while (*code == OP_ALT);   /* Loop for each alternative */  while (*code == OP_ALT);   /* Loop for each alternative */
# Line 6601  is_startline(const uschar *code, unsigne Line 7100  is_startline(const uschar *code, unsigne
7100  {  {
7101  do {  do {
7102     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
7103       NULL, 0, FALSE);       FALSE);
7104     register int op = *scode;     register int op = *scode;
7105    
7106     /* If we are at the start of a conditional assertion group, *both* the     /* If we are at the start of a conditional assertion group, *both* the
# Line 6628  do { Line 7127  do {
7127         scode += 1 + LINK_SIZE;         scode += 1 + LINK_SIZE;
7128         break;         break;
7129         }         }
7130       scode = first_significant_code(scode, NULL, 0, FALSE);       scode = first_significant_code(scode, FALSE);
7131       op = *scode;       op = *scode;
7132       }       }
7133    
7134     /* Non-capturing brackets */     /* Non-capturing brackets */
7135    
7136     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
7137           op == OP_SBRA || op == OP_SBRAPOS)
7138       {       {
7139       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
7140       }       }
7141    
7142     /* Capturing brackets */     /* Capturing brackets */
7143    
7144     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
7145                op == OP_SCBRA || op == OP_SCBRAPOS)
7146       {       {
7147       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
7148       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
# Line 6650  do { Line 7151  do {
7151    
7152     /* Other brackets */     /* Other brackets */
7153    
7154     else if (op == OP_ASSERT || op == OP_ONCE)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC)
7155       {       {
7156       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
7157       }       }
# Line 6665  do { Line 7166  do {
7166    
7167     /* Check for explicit circumflex */     /* Check for explicit circumflex */
7168    
7169     else if (op != OP_CIRC) return FALSE;     else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
7170    
7171     /* Move on to the next alternative */     /* Move on to the next alternative */
7172    
# Line 6691  we return that char, otherwise -1. Line 7192  we return that char, otherwise -1.
7192    
7193  Arguments:  Arguments:
7194    code       points to start of expression (the bracket)    code       points to start of expression (the bracket)
   options    pointer to the options (used to check casing changes)  
7195    inassert   TRUE if in an assertion    inassert   TRUE if in an assertion
7196    
7197  Returns:     -1 or the fixed first char  Returns:     -1 or the fixed first char
7198  */  */
7199    
7200  static int  static int
7201  find_firstassertedchar(const uschar *code, int *options, BOOL inassert)  find_firstassertedchar(const uschar *code, BOOL inassert)
7202  {  {
7203  register int c = -1;  register int c = -1;
7204  do {  do {
7205     int d;     int d;
7206     const uschar *scode =     int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
7207       first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);               *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? 2:0;
7208       const uschar *scode = first_significant_code(code + 1+LINK_SIZE + xl, TRUE);
7209     register int op = *scode;     register int op = *scode;
7210    
7211     switch(op)     switch(op)
# Line 6713  do { Line 7214  do {
7214       return -1;       return -1;
7215    
7216       case OP_BRA:       case OP_BRA:
7217         case OP_BRAPOS:
7218       case OP_CBRA:       case OP_CBRA:
7219         case OP_SCBRA:
7220         case OP_CBRAPOS:
7221         case OP_SCBRAPOS:
7222       case OP_ASSERT:       case OP_ASSERT:
7223       case OP_ONCE:       case OP_ONCE:
7224         case OP_ONCE_NC:
7225       case OP_COND:       case OP_COND:
7226       if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)       if ((d = find_firstassertedchar(scode, op == OP_ASSERT)) < 0)
7227         return -1;         return -1;
7228       if (c < 0) c = d; else if (c != d) return -1;       if (c < 0) c = d; else if (c != d) return -1;
7229       break;       break;
7230    
7231       case OP_EXACT:       /* Fall through */       case OP_EXACT:
7232       scode += 2;       scode += 2;
7233         /* Fall through */
7234    
7235       case OP_CHAR:       case OP_CHAR:
      case OP_CHARNC:  
7236       case OP_PLUS:       case OP_PLUS:
7237       case OP_MINPLUS:       case OP_MINPLUS:
7238       case OP_POSPLUS:       case OP_POSPLUS:
7239       if (!inassert) return -1;       if (!inassert) return -1;
7240       if (c < 0)       if (c < 0) c = scode[1];
7241         {         else if (c != scode[1]) return -1;
7242         c = scode[1];       break;
7243         if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;  
7244         }       case OP_EXACTI:
7245       else if (c != scode[1]) return -1;       scode += 2;
7246         /* Fall through */
7247    
7248         case OP_CHARI:
7249         case OP_PLUSI:
7250         case OP_MINPLUSI:
7251         case OP_POSPLUSI:
7252         if (!inassert) return -1;
7253         if (c < 0) c = scode[1] | REQ_CASELESS;
7254           else if (c != scode[1]) return -1;
7255       break;       break;
7256       }       }
7257    
# Line 6799  compile_data *cd = &compile_block; Line 7314  compile_data *cd = &compile_block;
7314  computing the amount of memory that is needed. Compiled items are thrown away  computing the amount of memory that is needed. Compiled items are thrown away
7315  as soon as possible, so that a fairly large buffer should be sufficient for  as soon as possible, so that a fairly large buffer should be sufficient for
7316  this purpose. The same space is used in the second phase for remembering where  this purpose. The same space is used in the second phase for remembering where
7317  to fill in forward references to subpatterns. */  to fill in forward references to subpatterns. That may overflow, in which case
7318    new memory is obtained from malloc(). */
7319    
7320  uschar cworkspace[COMPILE_WORK_SIZE];  uschar cworkspace[COMPILE_WORK_SIZE];
7321    
# Line 6859  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7375  while (ptr[skipatstart] == CHAR_LEFT_PAR
7375      { skipatstart += 7; options |= PCRE_UTF8; continue; }      { skipatstart += 7; options |= PCRE_UTF8; continue; }
7376    else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)    else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
7377      { skipatstart += 6; options |= PCRE_UCP; continue; }      { skipatstart += 6; options |= PCRE_UCP; continue; }
7378      else if (strncmp((char *)(ptr+skipatstart+2), STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
7379        { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
7380    
7381    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
7382      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
# Line 6885  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7403  while (ptr[skipatstart] == CHAR_LEFT_PAR
7403    
7404  utf8 = (options & PCRE_UTF8) != 0;  utf8 = (options & PCRE_UTF8) != 0;
7405    
7406  /* 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
7407    return of an error code from _pcre_valid_utf8() is a new feature, introduced in
7408    release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
7409    not used here. */
7410    
7411  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
7412  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
7413       (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)       (errorcode = _pcre_valid_utf8((USPTR)pattern, -1, erroroffset)) != 0)
7414    {    {
7415    errorcode = ERR44;    errorcode = ERR44;
7416    goto PCRE_EARLY_ERROR_RETURN2;    goto PCRE_EARLY_ERROR_RETURN2;
# Line 6914  if ((options & PCRE_UCP) != 0) Line 7435  if ((options & PCRE_UCP) != 0)
7435    
7436  /* Check validity of \R options. */  /* Check validity of \R options. */
7437    
7438  switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))  if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
7439         (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
7440    {    {
7441    case 0:    errorcode = ERR56;
7442    case PCRE_BSR_ANYCRLF:    goto PCRE_EARLY_ERROR_RETURN;
   case PCRE_BSR_UNICODE:  
   break;  
   default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;  
7443    }    }
7444    
7445  /* Handle different types of newline. The three bits give seven cases. The  /* Handle different types of newline. The three bits give seven cases. The
# Line 6986  cd->bracount = cd->final_bracount = 0; Line 7505  cd->bracount = cd->final_bracount = 0;
7505  cd->names_found = 0;  cd->names_found = 0;
7506  cd->name_entry_size = 0;  cd->name_entry_size = 0;
7507  cd->name_table = NULL;  cd->name_table = NULL;
 cd->start_workspace = cworkspace;  
7508  cd->start_code = cworkspace;  cd->start_code = cworkspace;
7509  cd->hwm = cworkspace;  cd->hwm = cworkspace;
7510    cd->start_workspace = cworkspace;
7511    cd->workspace_size = COMPILE_WORK_SIZE;
7512  cd->start_pattern = (const uschar *)pattern;  cd->start_pattern = (const uschar *)pattern;
7513  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
7514  cd->req_varyopt = 0;  cd->req_varyopt = 0;
# Line 7005  outside can help speed up starting point Line 7525  outside can help speed up starting point
7525  ptr += skipatstart;  ptr += skipatstart;
7526  code = cworkspace;  code = cworkspace;
7527  *code = OP_BRA;  *code = OP_BRA;
7528  (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,  (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
7529    &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,    FALSE, 0, 0, &firstbyte, &reqbyte, NULL, cd, &length);
   &length);  
7530  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
7531    
7532  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
# Line 7061  field; this time it's used for rememberi Line 7580  field; this time it's used for rememberi
7580  */  */
7581    
7582  cd->final_bracount = cd->bracount;  /* Save for checking forward references */  cd->final_bracount = cd->bracount;  /* Save for checking forward references */
7583    cd->assert_depth = 0;
7584  cd->bracount = 0;  cd->bracount = 0;
7585  cd->names_found = 0;  cd->names_found = 0;
7586  cd->name_table = (uschar *)re + re->name_table_offset;  cd->name_table = (uschar *)re + re->name_table_offset;
7587  codestart = cd->name_table + re->name_entry_size * re->name_count;  codestart = cd->name_table + re->name_entry_size * re->name_count;
7588  cd->start_code = codestart;  cd->start_code = codestart;
7589  cd->hwm = cworkspace;  cd->hwm = (uschar *)(cd->start_workspace);
7590  cd->req_varyopt = 0;  cd->req_varyopt = 0;
7591  cd->had_accept = FALSE;  cd->had_accept = FALSE;
7592  cd->check_lookbehind = FALSE;  cd->check_lookbehind = FALSE;
# Line 7079  of the function here. */ Line 7599  of the function here. */
7599  ptr = (const uschar *)pattern + skipatstart;  ptr = (const uschar *)pattern + skipatstart;
7600  code = (uschar *)codestart;  code = (uschar *)codestart;
7601  *code = OP_BRA;  *code = OP_BRA;
7602  (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,  (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
7603    &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);    &firstbyte, &reqbyte, NULL, cd, NULL);
7604  re->top_bracket = cd->bracount;  re->top_bracket = cd->bracount;
7605  re->top_backref = cd->top_backref;  re->top_backref = cd->top_backref;
7606  re->flags = cd->external_flags;  re->flags = cd->external_flags;
7607    
7608  if (cd->had_accept) reqbyte = -1;   /* Must disable after (*ACCEPT) */  if (cd->had_accept) reqbyte = REQ_NONE;   /* Must disable after (*ACCEPT) */
7609    
7610  /* If not reached end of pattern on success, there's an excess bracket. */  /* If not reached end of pattern on success, there's an excess bracket. */
7611    
# Line 7100  if debugging, leave the test till after Line 7620  if debugging, leave the test till after
7620  if (code - codestart > length) errorcode = ERR23;  if (code - codestart > length) errorcode = ERR23;
7621  #endif  #endif
7622    
7623  /* Fill in any forward references that are required. */  /* Fill in any forward references that are required. There may be repeated
7624    references; optimize for them, as searching a large regex takes time. */
7625    
7626  while (errorcode == 0 && cd->hwm > cworkspace)  if (cd->hwm > cd->start_workspace)
7627    {    {
7628    int offset, recno;    int prev_recno = -1;
7629    const uschar *groupptr;    const uschar *groupptr = NULL;
7630    cd->hwm -= LINK_SIZE;    while (errorcode == 0 && cd->hwm > cd->start_workspace)
7631    offset = GET(cd->hwm, 0);      {
7632    recno = GET(codestart, offset);      int offset, recno;
7633    groupptr = _pcre_find_bracket(codestart, utf8, recno);      cd->hwm -= LINK_SIZE;
7634    if (groupptr == NULL) errorcode = ERR53;      offset = GET(cd->hwm, 0);
7635      else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));      recno = GET(codestart, offset);
7636    }      if (recno != prev_recno)
7637          {
7638          groupptr = _pcre_find_bracket(codestart, utf8, recno);
7639          prev_recno = recno;
7640          }
7641        if (groupptr == NULL) errorcode = ERR53;
7642          else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
7643        }
7644      }
7645    
7646    /* If the workspace had to be expanded, free the new memory. */
7647    
7648    if (cd->workspace_size > COMPILE_WORK_SIZE)
7649      (pcre_free)((void *)cd->start_workspace);
7650    
7651  /* Give an error if there's back reference to a non-existent capturing  /* Give an error if there's back reference to a non-existent capturing
7652  subpattern. */  subpattern. */
# Line 7146  if (cd->check_lookbehind) Line 7680  if (cd->check_lookbehind)
7680        uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);        uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
7681        int end_op = *be;        int end_op = *be;
7682        *be = OP_END;        *be = OP_END;
7683        fixed_length = find_fixedlength(cc, re->options, TRUE, cd);        fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
7684            cd);
7685        *be = end_op;        *be = end_op;
7686        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
7687        if (fixed_length < 0)        if (fixed_length < 0)
7688          {          {
7689          errorcode = (fixed_length == -2)? ERR36 : ERR25;          errorcode = (fixed_length == -2)? ERR36 :
7690                        (fixed_length == -4)? ERR70 : ERR25;
7691          break;          break;
7692          }          }
7693        PUT(cc, 1, fixed_length);        PUT(cc, 1, fixed_length);
# Line 7185  start with ^. and also when all branches Line 7721  start with ^. and also when all branches
7721    
7722  if ((re->options & PCRE_ANCHORED) == 0)  if ((re->options & PCRE_ANCHORED) == 0)
7723    {    {
7724    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))  
7725      re->options |= PCRE_ANCHORED;      re->options |= PCRE_ANCHORED;
7726    else    else
7727      {      {
7728      if (firstbyte < 0)      if (firstbyte < 0)
7729        firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);        firstbyte = find_firstassertedchar(codestart, FALSE);
7730      if (firstbyte >= 0)   /* Remove caseless flag for non-caseable chars */      if (firstbyte >= 0)   /* Remove caseless flag for non-caseable chars */
7731        {        {
7732        int ch = firstbyte & 255;        int ch = firstbyte & 255;

Legend:
Removed from v.574  
changed lines
  Added in v.779

  ViewVC Help
Powered by ViewVC 1.1.5