/[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 807 by ph10, Sun Dec 18 10:03:38 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 1538  for (;;) Line 1655  for (;;)
1655      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1656      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
1657      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
1658      /* Fall through */      cc += _pcre_OP_lengths[*cc];
1659        break;
1660    
1661      /* Skip over things that don't match chars */      /* Skip over things that don't match chars */
1662    
1663      case OP_REVERSE:      case OP_MARK:
1664        case OP_PRUNE_ARG:
1665        case OP_SKIP_ARG:
1666        case OP_THEN_ARG:
1667        cc += cc[1] + _pcre_OP_lengths[*cc];
1668        break;
1669    
1670        case OP_CALLOUT:
1671        case OP_CIRC:
1672        case OP_CIRCM:
1673        case OP_CLOSE:
1674        case OP_COMMIT:
1675      case OP_CREF:      case OP_CREF:
     case OP_NCREF:  
     case OP_RREF:  
     case OP_NRREF:  
1676      case OP_DEF:      case OP_DEF:
1677      case OP_OPT:      case OP_DOLL:
1678      case OP_CALLOUT:      case OP_DOLLM:
     case OP_SOD:  
     case OP_SOM:  
     case OP_SET_SOM:  
1679      case OP_EOD:      case OP_EOD:
1680      case OP_EODN:      case OP_EODN:
1681      case OP_CIRC:      case OP_FAIL:
1682      case OP_DOLL:      case OP_NCREF:
1683        case OP_NRREF:
1684      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1685        case OP_PRUNE:
1686        case OP_REVERSE:
1687        case OP_RREF:
1688        case OP_SET_SOM:
1689        case OP_SKIP:
1690        case OP_SOD:
1691        case OP_SOM:
1692        case OP_THEN:
1693      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1694      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
1695      break;      break;
# Line 1565  for (;;) Line 1697  for (;;)
1697      /* Handle literal characters */      /* Handle literal characters */
1698    
1699      case OP_CHAR:      case OP_CHAR:
1700      case OP_CHARNC:      case OP_CHARI:
1701      case OP_NOT:      case OP_NOT:
1702        case OP_NOTI:
1703      branchlength++;      branchlength++;
1704      cc += 2;      cc += 2;
1705  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1706      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];  
1707  #endif  #endif
1708      break;      break;
1709    
# Line 1579  for (;;) Line 1711  for (;;)
1711      need to skip over a multibyte character in UTF8 mode.  */      need to skip over a multibyte character in UTF8 mode.  */
1712    
1713      case OP_EXACT:      case OP_EXACT:
1714        case OP_EXACTI:
1715        case OP_NOTEXACT:
1716        case OP_NOTEXACTI:
1717      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1718      cc += 4;      cc += 4;
1719  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1720      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];  
1721  #endif  #endif
1722      break;      break;
1723    
# Line 1600  for (;;) Line 1734  for (;;)
1734      cc += 2;      cc += 2;
1735      /* Fall through */      /* Fall through */
1736    
1737        case OP_HSPACE:
1738        case OP_VSPACE:
1739        case OP_NOT_HSPACE:
1740        case OP_NOT_VSPACE:
1741      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
1742      case OP_DIGIT:      case OP_DIGIT:
1743      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
# Line 1612  for (;;) Line 1750  for (;;)
1750      cc++;      cc++;
1751      break;      break;
1752    
1753      /* The single-byte matcher isn't allowed */      /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1754        otherwise \C is coded as OP_ALLANY. */
1755    
1756      case OP_ANYBYTE:      case OP_ANYBYTE:
1757      return -2;      return -2;
# Line 1631  for (;;) Line 1770  for (;;)
1770    
1771      switch (*cc)      switch (*cc)
1772        {        {
1773          case OP_CRPLUS:
1774          case OP_CRMINPLUS:
1775        case OP_CRSTAR:        case OP_CRSTAR:
1776        case OP_CRMINSTAR:        case OP_CRMINSTAR:
1777        case OP_CRQUERY:        case OP_CRQUERY:
# Line 1651  for (;;) Line 1792  for (;;)
1792    
1793      /* Anything else is variable length */      /* Anything else is variable length */
1794    
1795      default:      case OP_ANYNL:
1796        case OP_BRAMINZERO:
1797        case OP_BRAPOS:
1798        case OP_BRAPOSZERO:
1799        case OP_BRAZERO:
1800        case OP_CBRAPOS:
1801        case OP_EXTUNI:
1802        case OP_KETRMAX:
1803        case OP_KETRMIN:
1804        case OP_KETRPOS:
1805        case OP_MINPLUS:
1806        case OP_MINPLUSI:
1807        case OP_MINQUERY:
1808        case OP_MINQUERYI:
1809        case OP_MINSTAR:
1810        case OP_MINSTARI:
1811        case OP_MINUPTO:
1812        case OP_MINUPTOI:
1813        case OP_NOTMINPLUS:
1814        case OP_NOTMINPLUSI:
1815        case OP_NOTMINQUERY:
1816        case OP_NOTMINQUERYI:
1817        case OP_NOTMINSTAR:
1818        case OP_NOTMINSTARI:
1819        case OP_NOTMINUPTO:
1820        case OP_NOTMINUPTOI:
1821        case OP_NOTPLUS:
1822        case OP_NOTPLUSI:
1823        case OP_NOTPOSPLUS:
1824        case OP_NOTPOSPLUSI:
1825        case OP_NOTPOSQUERY:
1826        case OP_NOTPOSQUERYI:
1827        case OP_NOTPOSSTAR:
1828        case OP_NOTPOSSTARI:
1829        case OP_NOTPOSUPTO:
1830        case OP_NOTPOSUPTOI:
1831        case OP_NOTQUERY:
1832        case OP_NOTQUERYI:
1833        case OP_NOTSTAR:
1834        case OP_NOTSTARI:
1835        case OP_NOTUPTO:
1836        case OP_NOTUPTOI:
1837        case OP_PLUS:
1838        case OP_PLUSI:
1839        case OP_POSPLUS:
1840        case OP_POSPLUSI:
1841        case OP_POSQUERY:
1842        case OP_POSQUERYI:
1843        case OP_POSSTAR:
1844        case OP_POSSTARI:
1845        case OP_POSUPTO:
1846        case OP_POSUPTOI:
1847        case OP_QUERY:
1848        case OP_QUERYI:
1849        case OP_REF:
1850        case OP_REFI:
1851        case OP_SBRA:
1852        case OP_SBRAPOS:
1853        case OP_SCBRA:
1854        case OP_SCBRAPOS:
1855        case OP_SCOND:
1856        case OP_SKIPZERO:
1857        case OP_STAR:
1858        case OP_STARI:
1859        case OP_TYPEMINPLUS:
1860        case OP_TYPEMINQUERY:
1861        case OP_TYPEMINSTAR:
1862        case OP_TYPEMINUPTO:
1863        case OP_TYPEPLUS:
1864        case OP_TYPEPOSPLUS:
1865        case OP_TYPEPOSQUERY:
1866        case OP_TYPEPOSSTAR:
1867        case OP_TYPEPOSUPTO:
1868        case OP_TYPEQUERY:
1869        case OP_TYPESTAR:
1870        case OP_TYPEUPTO:
1871        case OP_UPTO:
1872        case OP_UPTOI:
1873      return -1;      return -1;
1874    
1875        /* Catch unrecognized opcodes so that when new ones are added they
1876        are not forgotten, as has happened in the past. */
1877    
1878        default:
1879        return -4;
1880      }      }
1881    }    }
1882  /* Control never gets here */  /* Control never gets here */
# Line 1685  _pcre_find_bracket(const uschar *code, B Line 1909  _pcre_find_bracket(const uschar *code, B
1909  for (;;)  for (;;)
1910    {    {
1911    register int c = *code;    register int c = *code;
1912    
1913    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1914    
1915    /* 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 1928  for (;;)
1928    
1929    /* Handle capturing bracket */    /* Handle capturing bracket */
1930    
1931    else if (c == OP_CBRA)    else if (c == OP_CBRA || c == OP_SCBRA ||
1932               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1933      {      {
1934      int n = GET2(code, 1+LINK_SIZE);      int n = GET2(code, 1+LINK_SIZE);
1935      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
# Line 1745  for (;;) Line 1971  for (;;)
1971        break;        break;
1972    
1973        case OP_THEN_ARG:        case OP_THEN_ARG:
1974        code += code[1+LINK_SIZE];        code += code[1];
1975        break;        break;
1976        }        }
1977    
# Line 1761  for (;;) Line 1987  for (;;)
1987      if (utf8) switch(c)      if (utf8) switch(c)
1988        {        {
1989        case OP_CHAR:        case OP_CHAR:
1990        case OP_CHARNC:        case OP_CHARI:
1991        case OP_EXACT:        case OP_EXACT:
1992          case OP_EXACTI:
1993        case OP_UPTO:        case OP_UPTO:
1994          case OP_UPTOI:
1995        case OP_MINUPTO:        case OP_MINUPTO:
1996          case OP_MINUPTOI:
1997        case OP_POSUPTO:        case OP_POSUPTO:
1998          case OP_POSUPTOI:
1999        case OP_STAR:        case OP_STAR:
2000          case OP_STARI:
2001        case OP_MINSTAR:        case OP_MINSTAR:
2002          case OP_MINSTARI:
2003        case OP_POSSTAR:        case OP_POSSTAR:
2004          case OP_POSSTARI:
2005        case OP_PLUS:        case OP_PLUS:
2006          case OP_PLUSI:
2007        case OP_MINPLUS:        case OP_MINPLUS:
2008          case OP_MINPLUSI:
2009        case OP_POSPLUS:        case OP_POSPLUS:
2010          case OP_POSPLUSI:
2011        case OP_QUERY:        case OP_QUERY:
2012          case OP_QUERYI:
2013        case OP_MINQUERY:        case OP_MINQUERY:
2014          case OP_MINQUERYI:
2015        case OP_POSQUERY:        case OP_POSQUERY:
2016          case OP_POSQUERYI:
2017        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2018        break;        break;
2019        }        }
# Line 1851  for (;;) Line 2090  for (;;)
2090        break;        break;
2091    
2092        case OP_THEN_ARG:        case OP_THEN_ARG:
2093        code += code[1+LINK_SIZE];        code += code[1];
2094        break;        break;
2095        }        }
2096    
# Line 1867  for (;;) Line 2106  for (;;)
2106      if (utf8) switch(c)      if (utf8) switch(c)
2107        {        {
2108        case OP_CHAR:        case OP_CHAR:
2109        case OP_CHARNC:        case OP_CHARI:
2110        case OP_EXACT:        case OP_EXACT:
2111          case OP_EXACTI:
2112        case OP_UPTO:        case OP_UPTO:
2113          case OP_UPTOI:
2114        case OP_MINUPTO:        case OP_MINUPTO:
2115          case OP_MINUPTOI:
2116        case OP_POSUPTO:        case OP_POSUPTO:
2117          case OP_POSUPTOI:
2118        case OP_STAR:        case OP_STAR:
2119          case OP_STARI:
2120        case OP_MINSTAR:        case OP_MINSTAR:
2121          case OP_MINSTARI:
2122        case OP_POSSTAR:        case OP_POSSTAR:
2123          case OP_POSSTARI:
2124        case OP_PLUS:        case OP_PLUS:
2125          case OP_PLUSI:
2126        case OP_MINPLUS:        case OP_MINPLUS:
2127          case OP_MINPLUSI:
2128        case OP_POSPLUS:        case OP_POSPLUS:
2129          case OP_POSPLUSI:
2130        case OP_QUERY:        case OP_QUERY:
2131          case OP_QUERYI:
2132        case OP_MINQUERY:        case OP_MINQUERY:
2133          case OP_MINQUERYI:
2134        case OP_POSQUERY:        case OP_POSQUERY:
2135          case OP_POSQUERYI:
2136        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
2137        break;        break;
2138        }        }
# Line 1919  could_be_empty_branch(const uschar *code Line 2171  could_be_empty_branch(const uschar *code
2171    compile_data *cd)    compile_data *cd)
2172  {  {
2173  register int c;  register int c;
2174  for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2175       code < endcode;       code < endcode;
2176       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2177    {    {
2178    const uschar *ccode;    const uschar *ccode;
2179    
# Line 1937  for (code = first_significant_code(code Line 2189  for (code = first_significant_code(code
2189      continue;      continue;
2190      }      }
2191    
   /* 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;  
     }  
   
2192    /* For a recursion/subroutine call, if its end has been reached, which    /* For a recursion/subroutine call, if its end has been reached, which
2193    implies a subroutine call, we can scan it. */    implies a backward reference subroutine call, we can scan it. If it's a
2194      forward reference subroutine call, we can't. To detect forward reference
2195      we have to scan up the list that is kept in the workspace. This function is
2196      called only when doing the real compile, not during the pre-compile that
2197      measures the size of the compiled pattern. */
2198    
2199    if (c == OP_RECURSE)    if (c == OP_RECURSE)
2200      {      {
2201      BOOL empty_branch = FALSE;      const uschar *scode;
2202      const uschar *scode = cd->start_code + GET(code, 1);      BOOL empty_branch;
2203    
2204        /* Test for forward reference */
2205    
2206        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2207          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2208    
2209        /* Not a forward reference, test for completed backward reference */
2210    
2211        empty_branch = FALSE;
2212        scode = cd->start_code + GET(code, 1);
2213      if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */      if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2214    
2215        /* Completed backwards reference */
2216    
2217      do      do
2218        {        {
2219        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 2224  for (code = first_significant_code(code
2224        scode += GET(scode, 1);        scode += GET(scode, 1);
2225        }        }
2226      while (*scode == OP_ALT);      while (*scode == OP_ALT);
2227    
2228      if (!empty_branch) return FALSE;  /* All branches are non-empty */      if (!empty_branch) return FALSE;  /* All branches are non-empty */
2229      continue;      continue;
2230      }      }
2231    
2232      /* Groups with zero repeats can of course be empty; skip them. */
2233    
2234      if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2235          c == OP_BRAPOSZERO)
2236        {
2237        code += _pcre_OP_lengths[c];
2238        do code += GET(code, 1); while (*code == OP_ALT);
2239        c = *code;
2240        continue;
2241        }
2242    
2243      /* A nested group that is already marked as "could be empty" can just be
2244      skipped. */
2245    
2246      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2247          c == OP_SCBRA || c == OP_SCBRAPOS)
2248        {
2249        do code += GET(code, 1); while (*code == OP_ALT);
2250        c = *code;
2251        continue;
2252        }
2253    
2254    /* For other groups, scan the branches. */    /* For other groups, scan the branches. */
2255    
2256    if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)    if (c == OP_BRA  || c == OP_BRAPOS ||
2257          c == OP_CBRA || c == OP_CBRAPOS ||
2258          c == OP_ONCE || c == OP_ONCE_NC ||
2259          c == OP_COND)
2260      {      {
2261      BOOL empty_branch;      BOOL empty_branch;
2262      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 2343  for (code = first_significant_code(code
2343      case OP_ALLANY:      case OP_ALLANY:
2344      case OP_ANYBYTE:      case OP_ANYBYTE:
2345      case OP_CHAR:      case OP_CHAR:
2346      case OP_CHARNC:      case OP_CHARI:
2347      case OP_NOT:      case OP_NOT:
2348        case OP_NOTI:
2349      case OP_PLUS:      case OP_PLUS:
2350      case OP_MINPLUS:      case OP_MINPLUS:
2351      case OP_POSPLUS:      case OP_POSPLUS:
# Line 2099  for (code = first_significant_code(code Line 2385  for (code = first_significant_code(code
2385      case OP_KET:      case OP_KET:
2386      case OP_KETRMAX:      case OP_KETRMAX:
2387      case OP_KETRMIN:      case OP_KETRMIN:
2388        case OP_KETRPOS:
2389      case OP_ALT:      case OP_ALT:
2390      return TRUE;      return TRUE;
2391    
# Line 2107  for (code = first_significant_code(code Line 2394  for (code = first_significant_code(code
2394    
2395  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2396      case OP_STAR:      case OP_STAR:
2397        case OP_STARI:
2398      case OP_MINSTAR:      case OP_MINSTAR:
2399        case OP_MINSTARI:
2400      case OP_POSSTAR:      case OP_POSSTAR:
2401        case OP_POSSTARI:
2402      case OP_QUERY:      case OP_QUERY:
2403        case OP_QUERYI:
2404      case OP_MINQUERY:      case OP_MINQUERY:
2405        case OP_MINQUERYI:
2406      case OP_POSQUERY:      case OP_POSQUERY:
2407        case OP_POSQUERYI:
2408      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2409      break;      break;
2410    
2411      case OP_UPTO:      case OP_UPTO:
2412        case OP_UPTOI:
2413      case OP_MINUPTO:      case OP_MINUPTO:
2414        case OP_MINUPTOI:
2415      case OP_POSUPTO:      case OP_POSUPTO:
2416        case OP_POSUPTOI:
2417      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2418      break;      break;
2419  #endif  #endif
# Line 2132  for (code = first_significant_code(code Line 2428  for (code = first_significant_code(code
2428      break;      break;
2429    
2430      case OP_THEN_ARG:      case OP_THEN_ARG:
2431      code += code[1+LINK_SIZE];      code += code[1];
2432      break;      break;
2433    
2434      /* 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 2451  return TRUE;
2451  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
2452  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,
2453  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.
2454    This function is called only during the real compile, not during the
2455    pre-compile.
2456    
2457  Arguments:  Arguments:
2458    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 2503  where Perl recognizes it as the POSIX cl
2503  "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,
2504  I think.  I think.
2505    
2506    A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2507    It seems that the appearance of a nested POSIX class supersedes an apparent
2508    external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2509    a digit.
2510    
2511    In Perl, unescaped square brackets may also appear as part of class names. For
2512    example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2513    [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2514    seem right at all. PCRE does not allow closing square brackets in POSIX class
2515    names.
2516    
2517  Arguments:  Arguments:
2518    ptr      pointer to the initial [    ptr      pointer to the initial [
2519    endptr   where to return the end pointer    endptr   where to return the end pointer
# Line 2219  int terminator;          /* Don't combin Line 2528  int terminator;          /* Don't combin
2528  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2529  for (++ptr; *ptr != 0; ptr++)  for (++ptr; *ptr != 0; ptr++)
2530    {    {
2531    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2532        ptr++;
2533      else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2534      else
2535      {      {
     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;  
2536      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2537        {        {
2538        *endptr = ptr;        *endptr = ptr;
2539        return TRUE;        return TRUE;
2540        }        }
2541        if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2542             (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2543              ptr[1] == CHAR_EQUALS_SIGN) &&
2544            check_posix_syntax(ptr, endptr))
2545          return FALSE;
2546      }      }
2547    }    }
2548  return FALSE;  return FALSE;
# Line 2532  if ((options & PCRE_EXTENDED) != 0) Line 2848  if ((options & PCRE_EXTENDED) != 0)
2848      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2849      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2850        {        {
2851        ptr++;        ptr++;
2852        while (*ptr != 0)        while (*ptr != 0)
2853          {          {
2854          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2855          ptr++;          ptr++;
2856  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2857          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2858  #endif  #endif
2859          }          }
# Line 2576  if ((options & PCRE_EXTENDED) != 0) Line 2892  if ((options & PCRE_EXTENDED) != 0)
2892      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2893      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2894        {        {
2895        ptr++;        ptr++;
2896        while (*ptr != 0)        while (*ptr != 0)
2897          {          {
2898          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2899          ptr++;          ptr++;
2900  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2901          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2902  #endif  #endif
2903          }          }
# Line 2609  if (next >= 0) switch(op_code) Line 2925  if (next >= 0) switch(op_code)
2925  #endif  #endif
2926    return c != next;    return c != next;
2927    
2928    /* 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
2929    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
2930    high-valued characters. */    high-valued characters. */
2931    
2932    case OP_CHARNC:    case OP_CHARI:
2933  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2934    GETCHARTEST(c, previous);    GETCHARTEST(c, previous);
2935  #else  #else
# Line 2636  if (next >= 0) switch(op_code) Line 2952  if (next >= 0) switch(op_code)
2952  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2953    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2954    
2955    /* 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
2956      opcodes are not used for multi-byte characters, because they are coded using
2957      an XCLASS instead. */
2958    
2959    case OP_NOT:    case OP_NOT:
2960      return (c = *previous) == next;
2961    
2962      case OP_NOTI:
2963    if ((c = *previous) == next) return TRUE;    if ((c = *previous) == next) return TRUE;
   if ((options & PCRE_CASELESS) == 0) return FALSE;  
2964  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2965    if (utf8)    if (utf8)
2966      {      {
# Line 2745  replaced by OP_PROP codes when PCRE_UCP Line 3065  replaced by OP_PROP codes when PCRE_UCP
3065  switch(op_code)  switch(op_code)
3066    {    {
3067    case OP_CHAR:    case OP_CHAR:
3068    case OP_CHARNC:    case OP_CHARI:
3069  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3070    GETCHARTEST(c, previous);    GETCHARTEST(c, previous);
3071  #else  #else
# Line 2932  Arguments: Line 3252  Arguments:
3252    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3253    reqbyteptr     set to the last literal character required, else < 0    reqbyteptr     set to the last literal character required, else < 0
3254    bcptr          points to current branch chain    bcptr          points to current branch chain
3255      cond_depth     conditional nesting depth
3256    cd             contains pointers to tables etc.    cd             contains pointers to tables etc.
3257    lengthptr      NULL during the real compile phase    lengthptr      NULL during the real compile phase
3258                   points to length accumulator during pre-compile phase                   points to length accumulator during pre-compile phase
# Line 2943  Returns:         TRUE on success Line 3264  Returns:         TRUE on success
3264  static BOOL  static BOOL
3265  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3266    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3267    compile_data *cd, int *lengthptr)    int cond_depth, compile_data *cd, int *lengthptr)
3268  {  {
3269  int repeat_type, op_type;  int repeat_type, op_type;
3270  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 3273  int greedy_default, greedy_non_default;
3273  int firstbyte, reqbyte;  int firstbyte, reqbyte;
3274  int zeroreqbyte, zerofirstbyte;  int zeroreqbyte, zerofirstbyte;
3275  int req_caseopt, reqvary, tempreqvary;  int req_caseopt, reqvary, tempreqvary;
3276  int options = *optionsptr;  int options = *optionsptr;               /* May change dynamically */
3277  int after_manual_callout = 0;  int after_manual_callout = 0;
3278  int length_prevgroup = 0;  int length_prevgroup = 0;
3279  register int c;  register int c;
# Line 2970  uschar *previous_callout = NULL; Line 3291  uschar *previous_callout = NULL;
3291  uschar *save_hwm = NULL;  uschar *save_hwm = NULL;
3292  uschar classbits[32];  uschar classbits[32];
3293    
3294    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3295    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3296    dynamically as we process the pattern. */
3297    
3298  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3299  BOOL class_utf8;  BOOL class_utf8;
3300  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8 = (options & PCRE_UTF8) != 0;
# Line 2978  uschar *class_utf8data_base; Line 3303  uschar *class_utf8data_base;
3303  uschar utf8_char[6];  uschar utf8_char[6];
3304  #else  #else
3305  BOOL utf8 = FALSE;  BOOL utf8 = FALSE;
 uschar *utf8_char = NULL;  
3306  #endif  #endif
3307    
3308  #ifdef PCRE_DEBUG  #ifdef PCRE_DEBUG
# Line 3029  for (;; ptr++) Line 3353  for (;; ptr++)
3353    int subfirstbyte;    int subfirstbyte;
3354    int terminator;    int terminator;
3355    int mclength;    int mclength;
3356      int tempbracount;
3357    uschar mcbuffer[8];    uschar mcbuffer[8];
3358    
3359    /* Get next byte in the pattern */    /* Get next byte in the pattern */
# Line 3053  for (;; ptr++) Line 3378  for (;; ptr++)
3378  #ifdef PCRE_DEBUG  #ifdef PCRE_DEBUG
3379      if (code > cd->hwm) cd->hwm = code;                 /* High water info */      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3380  #endif  #endif
3381      if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */      if (code > cd->start_workspace + cd->workspace_size -
3382            WORK_SIZE_SAFETY_MARGIN)                       /* Check for overrun */
3383        {        {
3384        *errorcodeptr = ERR52;        *errorcodeptr = ERR52;
3385        goto FAILED;        goto FAILED;
# Line 3076  for (;; ptr++) Line 3402  for (;; ptr++)
3402        }        }
3403    
3404      *lengthptr += (int)(code - last_code);      *lengthptr += (int)(code - last_code);
3405      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),
3406          c));
3407    
3408      /* 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
3409      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 3429  for (;; ptr++)
3429    /* 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
3430    reference list. */    reference list. */
3431    
3432    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)    else if (cd->hwm > cd->start_workspace + cd->workspace_size -
3433               WORK_SIZE_SAFETY_MARGIN)
3434      {      {
3435      *errorcodeptr = ERR52;      *errorcodeptr = ERR52;
3436      goto FAILED;      goto FAILED;
# Line 3150  for (;; ptr++) Line 3478  for (;; ptr++)
3478      previous_callout = NULL;      previous_callout = NULL;
3479      }      }
3480    
3481    /* In extended mode, skip white space and comments */    /* In extended mode, skip white space and comments. */
3482    
3483    if ((options & PCRE_EXTENDED) != 0)    if ((options & PCRE_EXTENDED) != 0)
3484      {      {
3485      if ((cd->ctypes[c] & ctype_space) != 0) continue;      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3486      if (c == CHAR_NUMBER_SIGN)      if (c == CHAR_NUMBER_SIGN)
3487        {        {
3488        ptr++;        ptr++;
3489        while (*ptr != 0)        while (*ptr != 0)
3490          {          {
3491          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3492          ptr++;          ptr++;
3493  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3494          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3495  #endif  #endif
3496          }          }
# Line 3209  for (;; ptr++) Line 3537  for (;; ptr++)
3537      the setting of any following char as a first character. */      the setting of any following char as a first character. */
3538    
3539      case CHAR_CIRCUMFLEX_ACCENT:      case CHAR_CIRCUMFLEX_ACCENT:
3540        previous = NULL;
3541      if ((options & PCRE_MULTILINE) != 0)      if ((options & PCRE_MULTILINE) != 0)
3542        {        {
3543        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3544          *code++ = OP_CIRCM;
3545        }        }
3546      previous = NULL;      else *code++ = OP_CIRC;
     *code++ = OP_CIRC;  
3547      break;      break;
3548    
3549      case CHAR_DOLLAR_SIGN:      case CHAR_DOLLAR_SIGN:
3550      previous = NULL;      previous = NULL;
3551      *code++ = OP_DOLL;      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3552      break;      break;
3553    
3554      /* 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 3684  for (;; ptr++)
3684    
3685        if (lengthptr != NULL)        if (lengthptr != NULL)
3686          {          {
3687          *lengthptr += class_utf8data - class_utf8data_base;          *lengthptr += (int)(class_utf8data - class_utf8data_base);
3688          class_utf8data = class_utf8data_base;          class_utf8data = class_utf8data_base;
3689          }          }
3690    
# Line 3491  for (;; ptr++) Line 3820  for (;; ptr++)
3820          if (*errorcodeptr != 0) goto FAILED;          if (*errorcodeptr != 0) goto FAILED;
3821    
3822          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 */
3823            else if (-c == ESC_N)            /* \N is not supported in a class */
3824              {
3825              *errorcodeptr = ERR71;
3826              goto FAILED;
3827              }
3828          else if (-c == ESC_Q)            /* Handle start of quoted string */          else if (-c == ESC_Q)            /* Handle start of quoted string */
3829            {            {
3830            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
# Line 3540  for (;; ptr++) Line 3874  for (;; ptr++)
3874              continue;              continue;
3875    
3876              /* 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
3877              if it was previously set by something earlier in the character              if it was previously set by something earlier in the character
3878              class. */              class. */
3879    
3880              case ESC_s:              case ESC_s:
3881              classbits[0] |= cbits[cbit_space];              classbits[0] |= cbits[cbit_space];
3882              classbits[1] |= cbits[cbit_space+1] & ~0x08;              classbits[1] |= cbits[cbit_space+1] & ~0x08;
3883              for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];              for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3884              continue;              continue;
3885    
# Line 3965  for (;; ptr++) Line 4299  for (;; ptr++)
4299    
4300      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
4301      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4302      operate on single-bytes only. This is an historical hangover. Maybe one day      operate on single-bytes characters only. This is an historical hangover.
4303      we can tidy these opcodes to handle multi-byte characters.      Maybe one day we can tidy these opcodes to handle multi-byte characters.
4304    
4305      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
4306      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.
4307      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
4308      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
4309      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
4310      reqbyte, save the previous value for reinstating. */      of reqbyte, save the previous value for reinstating. */
4311    
4312  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4313      if (class_charcount == 1 && !class_utf8 &&      if (class_charcount == 1 && !class_utf8 &&
# Line 3984  for (;; ptr++) Line 4318  for (;; ptr++)
4318        {        {
4319        zeroreqbyte = reqbyte;        zeroreqbyte = reqbyte;
4320    
4321        /* The OP_NOT opcode works on one-byte characters only. */        /* The OP_NOT[I] opcodes work on one-byte characters only. */
4322    
4323        if (negate_class)        if (negate_class)
4324          {          {
4325          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4326          zerofirstbyte = firstbyte;          zerofirstbyte = firstbyte;
4327          *code++ = OP_NOT;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4328          *code++ = class_lastchar;          *code++ = class_lastchar;
4329          break;          break;
4330          }          }
# Line 4049  for (;; ptr++) Line 4383  for (;; ptr++)
4383    
4384        /* Now fill in the complete length of the item */        /* Now fill in the complete length of the item */
4385    
4386        PUT(previous, 1, code - previous);        PUT(previous, 1, (int)(code - previous));
4387        break;   /* End of class handling */        break;   /* End of class handling */
4388        }        }
4389  #endif  #endif
# Line 4118  for (;; ptr++) Line 4452  for (;; ptr++)
4452      op_type = 0;                    /* Default single-char op codes */      op_type = 0;                    /* Default single-char op codes */
4453      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
4454    
4455      /* 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
4456      for an inserted OP_ONCE for the additional '+' extension. */      insert something before it. */
4457    
4458      tempcode = previous;      tempcode = previous;
4459    
# Line 4142  for (;; ptr++) Line 4476  for (;; ptr++)
4476        }        }
4477      else repeat_type = greedy_default;      else repeat_type = greedy_default;
4478    
4479        /* If previous was a recursion call, wrap it in atomic brackets so that
4480        previous becomes the atomic group. All recursions were so wrapped in the
4481        past, but it no longer happens for non-repeated recursions. In fact, the
4482        repeated ones could be re-implemented independently so as not to need this,
4483        but for the moment we rely on the code for repeating groups. */
4484    
4485        if (*previous == OP_RECURSE)
4486          {
4487          memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4488          *previous = OP_ONCE;
4489          PUT(previous, 1, 2 + 2*LINK_SIZE);
4490          previous[2 + 2*LINK_SIZE] = OP_KET;
4491          PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4492          code += 2 + 2 * LINK_SIZE;
4493          length_prevgroup = 3 + 3*LINK_SIZE;
4494    
4495          /* When actually compiling, we need to check whether this was a forward
4496          reference, and if so, adjust the offset. */
4497    
4498          if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4499            {
4500            int offset = GET(cd->hwm, -LINK_SIZE);
4501            if (offset == previous + 1 - cd->start_code)
4502              PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4503            }
4504          }
4505    
4506        /* Now handle repetition for the different types of item. */
4507    
4508      /* If previous was a character match, abolish the item and generate a      /* If previous was a character match, abolish the item and generate a
4509      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
4510      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
4511      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
4512      instead.  */      instead.  */
4513    
4514      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (*previous == OP_CHAR || *previous == OP_CHARI)
4515        {        {
4516          op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4517    
4518        /* 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
4519        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
4520        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 4525  for (;; ptr++)
4525          {          {
4526          uschar *lastchar = code - 1;          uschar *lastchar = code - 1;
4527          while((*lastchar & 0xc0) == 0x80) lastchar--;          while((*lastchar & 0xc0) == 0x80) lastchar--;
4528          c = code - lastchar;            /* Length of UTF-8 character */          c = (int)(code - lastchar);     /* Length of UTF-8 character */
4529          memcpy(utf8_char, lastchar, c); /* Save the char */          memcpy(utf8_char, lastchar, c); /* Save the char */
4530          c |= 0x80;                      /* Flag c as a length */          c |= 0x80;                      /* Flag c as a length */
4531          }          }
# Line 4194  for (;; ptr++) Line 4559  for (;; ptr++)
4559      /* If previous was a single negated character ([^a] or similar), we use      /* If previous was a single negated character ([^a] or similar), we use
4560      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-
4561      character repeats by setting opt_type to add a suitable offset into      character repeats by setting opt_type to add a suitable offset into
4562      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
4563      currently used only for single-byte chars. */      are currently used only for single-byte chars. */
4564    
4565      else if (*previous == OP_NOT)      else if (*previous == OP_NOT || *previous == OP_NOTI)
4566        {        {
4567        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4568        c = previous[1];        c = previous[1];
4569        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4570            repeat_max < 0 &&            repeat_max < 0 &&
# Line 4396  for (;; ptr++) Line 4761  for (;; ptr++)
4761  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4762               *previous == OP_XCLASS ||               *previous == OP_XCLASS ||
4763  #endif  #endif
4764               *previous == OP_REF)               *previous == OP_REF ||
4765                 *previous == OP_REFI)
4766        {        {
4767        if (repeat_max == 0)        if (repeat_max == 0)
4768          {          {
# Line 4430  for (;; ptr++) Line 4796  for (;; ptr++)
4796        }        }
4797    
4798      /* 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
4799      cases. */      cases. Note that at this point we can encounter only the "basic" bracket
4800        opcodes such as BRA and CBRA, as this is the place where they get converted
4801        into the more special varieties such as BRAPOS and SBRA. A test for >=
4802        OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4803        ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4804        repetition of assertions, but now it does, for Perl compatibility. */
4805    
4806      else if (*previous == OP_BRA  || *previous == OP_CBRA ||      else if (*previous >= OP_ASSERT && *previous <= OP_COND)
              *previous == OP_ONCE || *previous == OP_COND)  
4807        {        {
4808        register int i;        register int i;
       int ketoffset = 0;  
4809        int len = (int)(code - previous);        int len = (int)(code - previous);
4810        uschar *bralink = NULL;        uschar *bralink = NULL;
4811          uschar *brazeroptr = NULL;
4812    
4813        /* Repeating a DEFINE group is pointless */        /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4814          we just ignore the repeat. */
4815    
4816        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4817          {          goto END_REPEAT;
4818          *errorcodeptr = ERR55;  
4819          goto FAILED;        /* There is no sense in actually repeating assertions. The only potential
4820          }        use of repetition is in cases when the assertion is optional. Therefore,
4821          if the minimum is greater than zero, just ignore the repeat. If the
4822          maximum is not not zero or one, set it to 1. */
4823    
4824        /* If the maximum repeat count is unlimited, find the end of the bracket        if (*previous < OP_ONCE)    /* Assertion */
4825        by scanning through from the start, and compute the offset back to it          {
4826        from the current code pointer. There may be an OP_OPT setting following          if (repeat_min > 0) goto END_REPEAT;
4827        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);  
4828          }          }
4829    
4830        /* 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 4845  for (;; ptr++)
4845          **   goto END_REPEAT;          **   goto END_REPEAT;
4846          **   }          **   }
4847    
4848          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
4849          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
4850          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
4851          groups are referenced, we cannot do this selectively.          don't have a list of which groups are referenced, we cannot do this
4852            selectively.
4853    
4854          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
4855          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 4869  for (;; ptr++)
4869              *previous++ = OP_SKIPZERO;              *previous++ = OP_SKIPZERO;
4870              goto END_REPEAT;              goto END_REPEAT;
4871              }              }
4872              brazeroptr = previous;    /* Save for possessive optimizing */
4873            *previous++ = OP_BRAZERO + repeat_type;            *previous++ = OP_BRAZERO + repeat_type;
4874            }            }
4875    
# Line 4564  for (;; ptr++) Line 4932  for (;; ptr++)
4932              *lengthptr += delta;              *lengthptr += delta;
4933              }              }
4934    
4935            /* This is compiling for real */            /* This is compiling for real. If there is a set first byte for
4936              the group, and we have not yet set a "required byte", set it. Make
4937              sure there is enough workspace for copying forward references before
4938              doing the copy. */
4939    
4940            else            else
4941              {              {
4942              if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;              if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4943    
4944              for (i = 1; i < repeat_min; i++)              for (i = 1; i < repeat_min; i++)
4945                {                {
4946                uschar *hc;                uschar *hc;
4947                uschar *this_hwm = cd->hwm;                uschar *this_hwm = cd->hwm;
4948                memcpy(code, previous, len);                memcpy(code, previous, len);
4949    
4950                  while (cd->hwm > cd->start_workspace + cd->workspace_size -
4951                         WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
4952                    {
4953                    int save_offset = save_hwm - cd->start_workspace;
4954                    int this_offset = this_hwm - cd->start_workspace;
4955                    *errorcodeptr = expand_workspace(cd);
4956                    if (*errorcodeptr != 0) goto FAILED;
4957                    save_hwm = (uschar *)cd->start_workspace + save_offset;
4958                    this_hwm = (uschar *)cd->start_workspace + this_offset;
4959                    }
4960    
4961                for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)                for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4962                  {                  {
4963                  PUT(cd->hwm, 0, GET(hc, 0) + len);                  PUT(cd->hwm, 0, GET(hc, 0) + len);
# Line 4641  for (;; ptr++) Line 5025  for (;; ptr++)
5025              }              }
5026    
5027            memcpy(code, previous, len);            memcpy(code, previous, len);
5028    
5029              /* Ensure there is enough workspace for forward references before
5030              copying them. */
5031    
5032              while (cd->hwm > cd->start_workspace + cd->workspace_size -
5033                     WORK_SIZE_SAFETY_MARGIN - (this_hwm - save_hwm))
5034                {
5035                int save_offset = save_hwm - cd->start_workspace;
5036                int this_offset = this_hwm - cd->start_workspace;
5037                *errorcodeptr = expand_workspace(cd);
5038                if (*errorcodeptr != 0) goto FAILED;
5039                save_hwm = (uschar *)cd->start_workspace + save_offset;
5040                this_hwm = (uschar *)cd->start_workspace + this_offset;
5041                }
5042    
5043            for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)            for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
5044              {              {
5045              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 5065  for (;; ptr++)
5065            }            }
5066          }          }
5067    
5068        /* 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
5069        can't just offset backwards from the current code point, because we        ONCE brackets, that's all we need to do. However, possessively repeated
5070        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
5071        correct offset was computed above.        behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
5072          deal with possessive ONCEs specially.
5073    
5074        Then, when we are doing the actual compile phase, check to see whether        Otherwise, when we are doing the actual compile phase, check to see
5075        this group is a non-atomic one that could match an empty string. If so,        whether this group is one that could match an empty string. If so,
5076        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
5077        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
5078        atomic groups at runtime, but in a different way.] */        groups at runtime, but in a different way.]
5079    
5080          Then, if the quantifier was possessive and the bracket is not a
5081          conditional, we convert the BRA code to the POS form, and the KET code to
5082          KETRPOS. (It turns out to be convenient at runtime to detect this kind of
5083          subpattern at both the start and at the end.) The use of special opcodes
5084          makes it possible to reduce greatly the stack usage in pcre_exec(). If
5085          the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO.
5086    
5087          Then, if the minimum number of matches is 1 or 0, cancel the possessive
5088          flag so that the default action below, of wrapping everything inside
5089          atomic brackets, does not happen. When the minimum is greater than 1,
5090          there will be earlier copies of the group, and so we still have to wrap
5091          the whole thing. */
5092    
5093        else        else
5094          {          {
5095          uschar *ketcode = code - ketoffset;          uschar *ketcode = code - 1 - LINK_SIZE;
5096          uschar *bracode = ketcode - GET(ketcode, 1);          uschar *bracode = ketcode - GET(ketcode, 1);
5097          *ketcode = OP_KETRMAX + repeat_type;  
5098          if (lengthptr == NULL && *bracode != OP_ONCE)          /* Convert possessive ONCE brackets to non-capturing */
5099    
5100            if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) &&
5101                possessive_quantifier) *bracode = OP_BRA;
5102    
5103            /* For non-possessive ONCE brackets, all we need to do is to
5104            set the KET. */
5105    
5106            if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC)
5107              *ketcode = OP_KETRMAX + repeat_type;
5108    
5109            /* Handle non-ONCE brackets and possessive ONCEs (which have been
5110            converted to non-capturing above). */
5111    
5112            else
5113            {            {
5114            uschar *scode = bracode;            /* In the compile phase, check for empty string matching. */
5115            do  
5116              if (lengthptr == NULL)
5117              {              {
5118              if (could_be_empty_branch(scode, ketcode, utf8, cd))              uschar *scode = bracode;
5119                do
5120                {                {
5121                *bracode += OP_SBRA - OP_BRA;                if (could_be_empty_branch(scode, ketcode, utf8, cd))
5122                break;                  {
5123                    *bracode += OP_SBRA - OP_BRA;
5124                    break;
5125                    }
5126                  scode += GET(scode, 1);
5127                }                }
5128              scode += GET(scode, 1);              while (*scode == OP_ALT);
5129              }              }
5130            while (*scode == OP_ALT);  
5131              /* Handle possessive quantifiers. */
5132    
5133              if (possessive_quantifier)
5134                {
5135                /* For COND brackets, we wrap the whole thing in a possessively
5136                repeated non-capturing bracket, because we have not invented POS
5137                versions of the COND opcodes. Because we are moving code along, we
5138                must ensure that any pending recursive references are updated. */
5139    
5140                if (*bracode == OP_COND || *bracode == OP_SCOND)
5141                  {
5142                  int nlen = (int)(code - bracode);
5143                  *code = OP_END;
5144                  adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm);
5145                  memmove(bracode + 1+LINK_SIZE, bracode, nlen);
5146                  code += 1 + LINK_SIZE;
5147                  nlen += 1 + LINK_SIZE;
5148                  *bracode = OP_BRAPOS;
5149                  *code++ = OP_KETRPOS;
5150                  PUTINC(code, 0, nlen);
5151                  PUT(bracode, 1, nlen);
5152                  }
5153    
5154                /* For non-COND brackets, we modify the BRA code and use KETRPOS. */
5155    
5156                else
5157                  {
5158                  *bracode += 1;              /* Switch to xxxPOS opcodes */
5159                  *ketcode = OP_KETRPOS;
5160                  }
5161    
5162                /* If the minimum is zero, mark it as possessive, then unset the
5163                possessive flag when the minimum is 0 or 1. */
5164    
5165                if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
5166                if (repeat_min < 2) possessive_quantifier = FALSE;
5167                }
5168    
5169              /* Non-possessive quantifier */
5170    
5171              else *ketcode = OP_KETRMAX + repeat_type;
5172            }            }
5173          }          }
5174        }        }
# Line 4715  for (;; ptr++) Line 5189  for (;; ptr++)
5189        }        }
5190    
5191      /* If the character following a repeat is '+', or if certain optimization      /* If the character following a repeat is '+', or if certain optimization
5192      tests above succeeded, possessive_quantifier is TRUE. For some of the      tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
5193      simpler opcodes, there is an special alternative opcode for this. For      there are special alternative opcodes for this case. For anything else, we
5194      anything else, we wrap the entire repeated item inside OP_ONCE brackets.      wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
5195      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
5196      but the special opcodes can optimize it a bit. The repeated item starts at      special opcodes can optimize it.
5197      tempcode, not at previous, which might be the first part of a string whose  
5198      (former) last char we repeated.      Some (but not all) possessively repeated subpatterns have already been
5199        completely handled in the code just above. For them, possessive_quantifier
5200        is always FALSE at this stage.
5201    
5202        Note that the repeated item starts at tempcode, not at previous, which
5203        might be the first part of a string whose (former) last char we repeated.
5204    
5205      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
5206      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 5231  for (;; ptr++)
5231          case OP_QUERY: *tempcode = OP_POSQUERY; break;          case OP_QUERY: *tempcode = OP_POSQUERY; break;
5232          case OP_UPTO:  *tempcode = OP_POSUPTO; break;          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
5233    
5234          case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;          case OP_STARI:  *tempcode = OP_POSSTARI; break;
5235          case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;          case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
5236          case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;          case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
5237          case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;          case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
5238    
5239          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
5240          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
5241          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
5242          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
5243    
5244            case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
5245            case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
5246            case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
5247            case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
5248    
5249            case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
5250            case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
5251            case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
5252            case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
5253    
5254          /* Because we are moving code along, we must ensure that any          /* Because we are moving code along, we must ensure that any
5255          pending recursive references are updated. */          pending recursive references are updated. */
5256    
# Line 4815  for (;; ptr++) Line 5304  for (;; ptr++)
5304        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
5305        namelen = (int)(ptr - name);        namelen = (int)(ptr - name);
5306    
5307          /* It appears that Perl allows any characters whatsoever, other than
5308          a closing parenthesis, to appear in arguments, so we no longer insist on
5309          letters, digits, and underscores. */
5310    
5311        if (*ptr == CHAR_COLON)        if (*ptr == CHAR_COLON)
5312          {          {
5313          arg = ++ptr;          arg = ++ptr;
5314          while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
           || *ptr == '_') ptr++;  
5315          arglen = (int)(ptr - arg);          arglen = (int)(ptr - arg);
5316          }          }
5317    
# Line 4836  for (;; ptr++) Line 5328  for (;; ptr++)
5328          if (namelen == verbs[i].len &&          if (namelen == verbs[i].len &&
5329              strncmp((char *)name, vn, namelen) == 0)              strncmp((char *)name, vn, namelen) == 0)
5330            {            {
5331            /* Check for open captures before ACCEPT */            /* Check for open captures before ACCEPT and convert it to
5332              ASSERT_ACCEPT if in an assertion. */
5333    
5334            if (verbs[i].op == OP_ACCEPT)            if (verbs[i].op == OP_ACCEPT)
5335              {              {
5336              open_capitem *oc;              open_capitem *oc;
5337                if (arglen != 0)
5338                  {
5339                  *errorcodeptr = ERR59;
5340                  goto FAILED;
5341                  }
5342              cd->had_accept = TRUE;              cd->had_accept = TRUE;
5343              for (oc = cd->open_caps; oc != NULL; oc = oc->next)              for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5344                {                {
5345                *code++ = OP_CLOSE;                *code++ = OP_CLOSE;
5346                PUT2INC(code, 0, oc->number);                PUT2INC(code, 0, oc->number);
5347                }                }
5348                *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5349    
5350                /* Do not set firstbyte after *ACCEPT */
5351                if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5352              }              }
5353    
5354            /* Handle the cases with/without an argument */            /* Handle other cases with/without an argument */
5355    
5356            if (arglen == 0)            else if (arglen == 0)
5357              {              {
5358              if (verbs[i].op < 0)   /* Argument is mandatory */              if (verbs[i].op < 0)   /* Argument is mandatory */
5359                {                {
# Line 4859  for (;; ptr++) Line 5361  for (;; ptr++)
5361                goto FAILED;                goto FAILED;
5362                }                }
5363              *code = verbs[i].op;              *code = verbs[i].op;
5364              if (*code++ == OP_THEN)              if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN;
               {  
               PUT(code, 0, code - bcptr->current_branch - 1);  
               code += LINK_SIZE;  
               }  
5365              }              }
5366    
5367            else            else
# Line 4874  for (;; ptr++) Line 5372  for (;; ptr++)
5372                goto FAILED;                goto FAILED;
5373                }                }
5374              *code = verbs[i].op_arg;              *code = verbs[i].op_arg;
5375              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;  
               }  
5376              *code++ = arglen;              *code++ = arglen;
5377              memcpy(code, arg, arglen);              memcpy(code, arg, arglen);
5378              code += arglen;              code += arglen;
# Line 5140  for (;; ptr++) Line 5634  for (;; ptr++)
5634          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5635          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
5636          bravalue = OP_ASSERT;          bravalue = OP_ASSERT;
5637            cd->assert_depth += 1;
5638          ptr++;          ptr++;
5639          break;          break;
5640    
# Line 5154  for (;; ptr++) Line 5649  for (;; ptr++)
5649            continue;            continue;
5650            }            }
5651          bravalue = OP_ASSERT_NOT;          bravalue = OP_ASSERT_NOT;
5652            cd->assert_depth += 1;
5653          break;          break;
5654    
5655    
# Line 5163  for (;; ptr++) Line 5659  for (;; ptr++)
5659            {            {
5660            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
5661            bravalue = OP_ASSERTBACK;            bravalue = OP_ASSERTBACK;
5662              cd->assert_depth += 1;
5663            ptr += 2;            ptr += 2;
5664            break;            break;
5665    
5666            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
5667            bravalue = OP_ASSERTBACK_NOT;            bravalue = OP_ASSERTBACK_NOT;
5668              cd->assert_depth += 1;
5669            ptr += 2;            ptr += 2;
5670            break;            break;
5671    
# Line 5189  for (;; ptr++) Line 5687  for (;; ptr++)
5687    
5688          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5689          case CHAR_C:                 /* Callout - may be followed by digits; */          case CHAR_C:                 /* Callout - may be followed by digits; */
5690          previous_callout = code;  /* Save for later completion */          previous_callout = code;     /* Save for later completion */
5691          after_manual_callout = 1; /* Skip one item before completing */          after_manual_callout = 1;    /* Skip one item before completing */
5692          *code++ = OP_CALLOUT;          *code++ = OP_CALLOUT;
5693            {            {
5694            int n = 0;            int n = 0;
# Line 5373  for (;; ptr++) Line 5871  for (;; ptr++)
5871          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5872          namelen = (int)(ptr - name);          namelen = (int)(ptr - name);
5873    
5874          /* 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
5875          reference number. */          a dummy reference number, because it was not used in the first pass.
5876            However, with the change of recursive back references to be atomic,
5877            we have to look for the number so that this state can be identified, as
5878            otherwise the incorrect length is computed. If it's not a backwards
5879            reference, the dummy number will do. */
5880    
5881          if (lengthptr != NULL)          if (lengthptr != NULL)
5882            {            {
5883              const uschar *temp;
5884    
5885            if (namelen == 0)            if (namelen == 0)
5886              {              {
5887              *errorcodeptr = ERR62;              *errorcodeptr = ERR62;
# Line 5393  for (;; ptr++) Line 5897  for (;; ptr++)
5897              *errorcodeptr = ERR48;              *errorcodeptr = ERR48;
5898              goto FAILED;              goto FAILED;
5899              }              }
5900            recno = 0;  
5901              /* The name table does not exist in the first pass, so we cannot
5902              do a simple search as in the code below. Instead, we have to scan the
5903              pattern to find the number. It is important that we scan it only as
5904              far as we have got because the syntax of named subpatterns has not
5905              been checked for the rest of the pattern, and find_parens() assumes
5906              correct syntax. In any case, it's a waste of resources to scan
5907              further. We stop the scan at the current point by temporarily
5908              adjusting the value of cd->endpattern. */
5909    
5910              temp = cd->end_pattern;
5911              cd->end_pattern = ptr;
5912              recno = find_parens(cd, name, namelen,
5913                (options & PCRE_EXTENDED) != 0, utf8);
5914              cd->end_pattern = temp;
5915              if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
5916            }            }
5917    
5918          /* 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 6056  for (;; ptr++)
6056    
6057                /* 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
6058                offset below, what it actually inserted is the reference number                offset below, what it actually inserted is the reference number
6059                of the group. */                of the group. Then remember the forward reference. */
6060    
6061                called = cd->start_code + recno;                called = cd->start_code + recno;
6062                PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));                if (cd->hwm >= cd->start_workspace + cd->workspace_size -
6063                      WORK_SIZE_SAFETY_MARGIN)
6064                    {
6065                    *errorcodeptr = expand_workspace(cd);
6066                    if (*errorcodeptr != 0) goto FAILED;
6067                    }
6068                  PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
6069                }                }
6070    
6071              /* If not a forward reference, and the subpattern is still open,              /* If not a forward reference, and the subpattern is still open,
6072              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
6073              recursion that could loop for ever, and diagnose that case. */              recursion that could loop for ever, and diagnose that case. We
6074                must not, however, do this check if we are in a conditional
6075                subpattern because the condition might be testing for recursion in
6076                a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
6077                Forever loops are also detected at runtime, so those that occur in
6078                conditional subpatterns will be picked up then. */
6079    
6080              else if (GET(called, 1) == 0 &&              else if (GET(called, 1) == 0 && cond_depth <= 0 &&
6081                       could_be_empty(called, code, bcptr, utf8, cd))                       could_be_empty(called, code, bcptr, utf8, cd))
6082                {                {
6083                *errorcodeptr = ERR40;                *errorcodeptr = ERR40;
# Line 5555  for (;; ptr++) Line 6085  for (;; ptr++)
6085                }                }
6086              }              }
6087    
6088            /* Insert the recursion/subroutine item, automatically wrapped inside            /* Insert the recursion/subroutine item. It does not have a set first
6089            "once" brackets. Set up a "previous group" length so that a            byte (relevant if it is repeated, because it will then be wrapped
6090            subsequent quantifier will work. */            with ONCE brackets). */
   
           *code = OP_ONCE;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
6091    
6092            *code = OP_RECURSE;            *code = OP_RECURSE;
6093            PUT(code, 1, (int)(called - cd->start_code));            PUT(code, 1, (int)(called - cd->start_code));
6094            code += 1 + LINK_SIZE;            code += 1 + LINK_SIZE;
6095              groupsetfirstbyte = FALSE;
           *code = OP_KET;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
   
           length_prevgroup = 3 + 3*LINK_SIZE;  
6096            }            }
6097    
6098          /* Can't determine a first byte now */          /* Can't determine a first byte now */
# Line 5632  for (;; ptr++) Line 6153  for (;; ptr++)
6153          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
6154          both phases.          both phases.
6155    
6156          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
6157          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. */  
6158    
6159          if (*ptr == CHAR_RIGHT_PARENTHESIS)          if (*ptr == CHAR_RIGHT_PARENTHESIS)
6160            {            {
# Line 5645  for (;; ptr++) Line 6165  for (;; ptr++)
6165              }              }
6166            else            else
6167              {              {
             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))  
               {  
               *code++ = OP_OPT;  
               *code++ = newoptions & PCRE_IMS;  
               }  
6168              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
6169              greedy_non_default = greedy_default ^ 1;              greedy_non_default = greedy_default ^ 1;
6170              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
6171              }              }
6172    
6173            /* Change options at this level, and pass them back for use            /* Change options at this level, and pass them back for use
6174            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). */  
6175    
6176            *optionsptr = options = newoptions;            *optionsptr = options = newoptions;
6177            previous = NULL;       /* This item can't be repeated */            previous = NULL;       /* This item can't be repeated */
# Line 5694  for (;; ptr++) Line 6207  for (;; ptr++)
6207        skipbytes = 2;        skipbytes = 2;
6208        }        }
6209    
6210      /* Process nested bracketed regex. Assertions may not be repeated, but      /* Process nested bracketed regex. Assertions used not to be repeatable,
6211      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
6212      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
6213      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. */  
6214    
6215      previous = (bravalue >= OP_ONCE)? code : NULL;      previous = code;                      /* For handling repetition */
6216      *code = bravalue;      *code = bravalue;
6217      tempcode = code;      tempcode = code;
6218      tempreqvary = cd->req_varyopt;     /* Save value before bracket */      tempreqvary = cd->req_varyopt;        /* Save value before bracket */
6219      length_prevgroup = 0;              /* Initialize for pre-compile phase */      tempbracount = cd->bracount;          /* Save value before bracket */
6220        length_prevgroup = 0;                 /* Initialize for pre-compile phase */
6221    
6222      if (!compile_regex(      if (!compile_regex(
6223           newoptions,                   /* The complete new option state */           newoptions,                      /* The complete new option state */
6224           options & PCRE_IMS,           /* The previous ims option state */           &tempcode,                       /* Where to put code (updated) */
6225           &tempcode,                    /* Where to put code (updated) */           &ptr,                            /* Input pointer (updated) */
6226           &ptr,                         /* Input pointer (updated) */           errorcodeptr,                    /* Where to put an error message */
          errorcodeptr,                 /* Where to put an error message */  
6227           (bravalue == OP_ASSERTBACK ||           (bravalue == OP_ASSERTBACK ||
6228            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
6229           reset_bracount,               /* True if (?| group */           reset_bracount,                  /* True if (?| group */
6230           skipbytes,                    /* Skip over bracket number */           skipbytes,                       /* Skip over bracket number */
6231           &subfirstbyte,                /* For possible first char */           cond_depth +
6232           &subreqbyte,                  /* For possible last char */             ((bravalue == OP_COND)?1:0),   /* Depth of condition subpatterns */
6233           bcptr,                        /* Current branch chain */           &subfirstbyte,                   /* For possible first char */
6234           cd,                           /* Tables block */           &subreqbyte,                     /* For possible last char */
6235           (lengthptr == NULL)? NULL :   /* Actual compile phase */           bcptr,                           /* Current branch chain */
6236             &length_prevgroup           /* Pre-compile phase */           cd,                              /* Tables block */
6237             (lengthptr == NULL)? NULL :      /* Actual compile phase */
6238               &length_prevgroup              /* Pre-compile phase */
6239           ))           ))
6240        goto FAILED;        goto FAILED;
6241    
6242        /* If this was an atomic group and there are no capturing groups within it,
6243        generate OP_ONCE_NC instead of OP_ONCE. */
6244    
6245        if (bravalue == OP_ONCE && cd->bracount <= tempbracount)
6246          *code = OP_ONCE_NC;
6247    
6248        if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
6249          cd->assert_depth -= 1;
6250    
6251      /* 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
6252      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.
6253      and any option resetting that may follow it. The pattern pointer (ptr)      The pattern pointer (ptr) is on the bracket.
     is on the bracket. */  
6254    
6255      /* 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
6256      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
6257      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
6258      not be available. */      not be available. */
# Line 5795  for (;; ptr++) Line 6317  for (;; ptr++)
6317          goto FAILED;          goto FAILED;
6318          }          }
6319        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6320        *code++ = OP_BRA;        code++;   /* This already contains bravalue */
6321        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
6322        *code++ = OP_KET;        *code++ = OP_KET;
6323        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
# Line 5929  for (;; ptr++) Line 6451  for (;; ptr++)
6451    
6452          if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)          if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
6453            {            {
6454            BOOL isnumber = TRUE;            BOOL is_a_number = TRUE;
6455            for (p = ptr + 1; *p != 0 && *p != terminator; p++)            for (p = ptr + 1; *p != 0 && *p != terminator; p++)
6456              {              {
6457              if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;              if ((cd->ctypes[*p] & ctype_digit) == 0) is_a_number = FALSE;
6458              if ((cd->ctypes[*p] & ctype_word) == 0) break;              if ((cd->ctypes[*p] & ctype_word) == 0) break;
6459              }              }
6460            if (*p != terminator)            if (*p != terminator)
# Line 5940  for (;; ptr++) Line 6462  for (;; ptr++)
6462              *errorcodeptr = ERR57;              *errorcodeptr = ERR57;
6463              break;              break;
6464              }              }
6465            if (isnumber)            if (is_a_number)
6466              {              {
6467              ptr++;              ptr++;
6468              goto HANDLE_NUMERICAL_RECURSION;              goto HANDLE_NUMERICAL_RECURSION;
# Line 5963  for (;; ptr++) Line 6485  for (;; ptr++)
6485          }          }
6486    
6487        /* \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).
6488        We also support \k{name} (.NET syntax) */        We also support \k{name} (.NET syntax).  */
6489    
6490        if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||        if (-c == ESC_k)
           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))  
6491          {          {
6492            if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6493              ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6494              {
6495              *errorcodeptr = ERR69;
6496              break;
6497              }
6498          is_recurse = FALSE;          is_recurse = FALSE;
6499          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6500            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
# Line 5987  for (;; ptr++) Line 6514  for (;; ptr++)
6514          HANDLE_REFERENCE:    /* Come here from named backref handling */          HANDLE_REFERENCE:    /* Come here from named backref handling */
6515          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6516          previous = code;          previous = code;
6517          *code++ = OP_REF;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6518          PUT2INC(code, 0, recno);          PUT2INC(code, 0, recno);
6519          cd->backref_map |= (recno < 32)? (1 << recno) : 1;          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6520          if (recno > cd->top_backref) cd->top_backref = recno;          if (recno > cd->top_backref) cd->top_backref = recno;
# Line 6047  for (;; ptr++) Line 6574  for (;; ptr++)
6574            }            }
6575          else          else
6576  #endif  #endif
6577            /* In non-UTF-8 mode, we turn \C into OP_ALLANY instead of OP_ANYBYTE
6578            so that it works in DFA mode and in lookbehinds. */
6579    
6580            {            {
6581            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;            previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6582            *code++ = -c;            *code++ = (!utf8 && c == -ESC_C)? OP_ALLANY : -c;
6583            }            }
6584          }          }
6585        continue;        continue;
# Line 6095  for (;; ptr++) Line 6625  for (;; ptr++)
6625    
6626      ONE_CHAR:      ONE_CHAR:
6627      previous = code;      previous = code;
6628      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6629      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6630    
6631      /* Remember if \r or \n were seen */      /* Remember if \r or \n were seen */
# Line 6124  for (;; ptr++) Line 6654  for (;; ptr++)
6654        else firstbyte = reqbyte = REQ_NONE;        else firstbyte = reqbyte = REQ_NONE;
6655        }        }
6656    
6657      /* 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
6658      1 or the matching is caseful. */      1 or the matching is caseful. */
6659    
6660      else      else
# Line 6159  return FALSE; Line 6689  return FALSE;
6689  /* 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
6690  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
6691  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.  
   
6692  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
6693  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
6694  value of lengthptr distinguishes the two phases.  value of lengthptr distinguishes the two phases.
6695    
6696  Arguments:  Arguments:
6697    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  
6698    codeptr        -> the address of the current code pointer    codeptr        -> the address of the current code pointer
6699    ptrptr         -> the address of the current pattern pointer    ptrptr         -> the address of the current pattern pointer
6700    errorcodeptr   -> pointer to error code variable    errorcodeptr   -> pointer to error code variable
6701    lookbehind     TRUE if this is a lookbehind assertion    lookbehind     TRUE if this is a lookbehind assertion
6702    reset_bracount TRUE to reset the count for each branch    reset_bracount TRUE to reset the count for each branch
6703    skipbytes      skip this many bytes at start (for brackets and OP_COND)    skipbytes      skip this many bytes at start (for brackets and OP_COND)
6704      cond_depth     depth of nesting for conditional subpatterns
6705    firstbyteptr   place to put the first required character, or a negative number    firstbyteptr   place to put the first required character, or a negative number
6706    reqbyteptr     place to put the last required character, or a negative number    reqbyteptr     place to put the last required character, or a negative number
6707    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 6713  Returns:         TRUE on success
6713  */  */
6714    
6715  static BOOL  static BOOL
6716  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,  compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6717    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6718    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,    int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6719    int *lengthptr)    compile_data *cd, int *lengthptr)
6720  {  {
6721  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
6722  uschar *code = *codeptr;  uschar *code = *codeptr;
# Line 6205  int branchfirstbyte, branchreqbyte; Line 6730  int branchfirstbyte, branchreqbyte;
6730  int length;  int length;
6731  int orig_bracount;  int orig_bracount;
6732  int max_bracount;  int max_bracount;
 int old_external_options = cd->external_options;  
6733  branch_chain bc;  branch_chain bc;
6734    
6735  bc.outer = bcptr;  bc.outer = bcptr;
# Line 6229  pre-compile phase to find out whether an Line 6753  pre-compile phase to find out whether an
6753    
6754  /* 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
6755  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
6756  detect groups that contain recursive back references to themselves. */  detect groups that contain recursive back references to themselves. Note that
6757    only OP_CBRA need be tested here; changing this opcode to one of its variants,
6758    e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6759    
6760  if (*code == OP_CBRA)  if (*code == OP_CBRA)
6761    {    {
# Line 6255  for (;;) Line 6781  for (;;)
6781    
6782    if (reset_bracount) cd->bracount = orig_bracount;    if (reset_bracount) cd->bracount = orig_bracount;
6783    
   /* 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;  
     }  
   
6784    /* Set up dummy OP_REVERSE if lookbehind assertion */    /* Set up dummy OP_REVERSE if lookbehind assertion */
6785    
6786    if (lookbehind)    if (lookbehind)
# Line 6278  for (;;) Line 6795  for (;;)
6795    into the length. */    into the length. */
6796    
6797    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6798          &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))          &branchreqbyte, &bc, cond_depth, cd,
6799            (lengthptr == NULL)? NULL : &length))
6800      {      {
6801      *ptrptr = ptr;      *ptrptr = ptr;
6802      return FALSE;      return FALSE;
6803      }      }
6804    
   /* 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;  
   
6805    /* Keep the highest bracket count in case (?| was used and some branch    /* Keep the highest bracket count in case (?| was used and some branch
6806    has fewer than the rest. */    has fewer than the rest. */
6807    
# Line 6353  for (;;) Line 6862  for (;;)
6862        {        {
6863        int fixed_length;        int fixed_length;
6864        *code = OP_END;        *code = OP_END;
6865        fixed_length = find_fixedlength(last_branch, options, FALSE, cd);        fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
6866            FALSE, cd);
6867        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
6868        if (fixed_length == -3)        if (fixed_length == -3)
6869          {          {
# Line 6361  for (;;) Line 6871  for (;;)
6871          }          }
6872        else if (fixed_length < 0)        else if (fixed_length < 0)
6873          {          {
6874          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;          *errorcodeptr = (fixed_length == -2)? ERR36 :
6875                            (fixed_length == -4)? ERR70: ERR25;
6876          *ptrptr = ptr;          *ptrptr = ptr;
6877          return FALSE;          return FALSE;
6878          }          }
# Line 6374  for (;;) Line 6885  for (;;)
6885    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
6886    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
6887    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
6888    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. */  
6889    
6890    if (*ptr != CHAR_VERTICAL_LINE)    if (*ptr != CHAR_VERTICAL_LINE)
6891      {      {
# Line 6420  for (;;) Line 6929  for (;;)
6929        cd->open_caps = cd->open_caps->next;        cd->open_caps = cd->open_caps->next;
6930        }        }
6931    
     /* Reset options if needed. */  
   
     if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)  
       {  
       *code++ = OP_OPT;  
       *code++ = oldims;  
       length += 2;  
       }  
   
6932      /* Retain the highest bracket number, in case resetting was used. */      /* Retain the highest bracket number, in case resetting was used. */
6933    
6934      cd->bracount = max_bracount;      cd->bracount = max_bracount;
# Line 6488  for (;;) Line 6988  for (;;)
6988  /* 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
6989  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
6990  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
6991  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
6992  counts, since OP_CIRC can match in the middle.  be found, because ^ generates OP_CIRCM in that mode.
6993    
6994  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.
6995  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 7010  of the more common cases more precisely.
7010    
7011  Arguments:  Arguments:
7012    code           points to start of expression (the bracket)    code           points to start of expression (the bracket)
   options        points to the options setting  
7013    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
7014                    handles up to substring 31; after that we just have to take                    handles up to substring 31; after that we just have to take
7015                    the less precise approach                    the less precise approach
# Line 6520  Returns:     TRUE or FALSE Line 7019  Returns:     TRUE or FALSE
7019  */  */
7020    
7021  static BOOL  static BOOL
7022  is_anchored(register const uschar *code, int *options, unsigned int bracket_map,  is_anchored(register const uschar *code, unsigned int bracket_map,
7023    unsigned int backref_map)    unsigned int backref_map)
7024  {  {
7025  do {  do {
7026     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
7027       options, PCRE_MULTILINE, FALSE);       FALSE);
7028     register int op = *scode;     register int op = *scode;
7029    
7030     /* Non-capturing brackets */     /* Non-capturing brackets */
7031    
7032     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
7033           op == OP_SBRA || op == OP_SBRAPOS)
7034       {       {
7035       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
7036       }       }
7037    
7038     /* Capturing brackets */     /* Capturing brackets */
7039    
7040     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
7041                op == OP_SCBRA || op == OP_SCBRAPOS)
7042       {       {
7043       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
7044       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
7045       if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;       if (!is_anchored(scode, new_map, backref_map)) return FALSE;
7046       }       }
7047    
7048     /* Other brackets */     /* Other brackets */
7049    
7050     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC ||
7051                op == OP_COND)
7052       {       {
7053       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
7054       }       }
7055    
7056     /* .* 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 7065  do {
7065    
7066     /* Check for explicit anchoring */     /* Check for explicit anchoring */
7067    
7068     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;  
7069     code += GET(code, 1);     code += GET(code, 1);
7070     }     }
7071  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 7101  is_startline(const uschar *code, unsigne
7101  {  {
7102  do {  do {
7103     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
7104       NULL, 0, FALSE);       FALSE);
7105     register int op = *scode;     register int op = *scode;
7106    
7107     /* 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 7128  do {
7128         scode += 1 + LINK_SIZE;         scode += 1 + LINK_SIZE;
7129         break;         break;
7130         }         }
7131       scode = first_significant_code(scode, NULL, 0, FALSE);       scode = first_significant_code(scode, FALSE);
7132       op = *scode;       op = *scode;
7133       }       }
7134    
7135     /* Non-capturing brackets */     /* Non-capturing brackets */
7136    
7137     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
7138           op == OP_SBRA || op == OP_SBRAPOS)
7139       {       {
7140       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
7141       }       }
7142    
7143     /* Capturing brackets */     /* Capturing brackets */
7144    
7145     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
7146                op == OP_SCBRA || op == OP_SCBRAPOS)
7147       {       {
7148       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
7149       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
# Line 6650  do { Line 7152  do {
7152    
7153     /* Other brackets */     /* Other brackets */
7154    
7155     else if (op == OP_ASSERT || op == OP_ONCE)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_ONCE_NC)
7156       {       {
7157       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
7158       }       }
# Line 6665  do { Line 7167  do {
7167    
7168     /* Check for explicit circumflex */     /* Check for explicit circumflex */
7169    
7170     else if (op != OP_CIRC) return FALSE;     else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
7171    
7172     /* Move on to the next alternative */     /* Move on to the next alternative */
7173    
# Line 6691  we return that char, otherwise -1. Line 7193  we return that char, otherwise -1.
7193    
7194  Arguments:  Arguments:
7195    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)  
7196    inassert   TRUE if in an assertion    inassert   TRUE if in an assertion
7197    
7198  Returns:     -1 or the fixed first char  Returns:     -1 or the fixed first char
7199  */  */
7200    
7201  static int  static int
7202  find_firstassertedchar(const uschar *code, int *options, BOOL inassert)  find_firstassertedchar(const uschar *code, BOOL inassert)
7203  {  {
7204  register int c = -1;  register int c = -1;
7205  do {  do {
7206     int d;     int d;
7207     const uschar *scode =     int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
7208       first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);               *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? 2:0;
7209       const uschar *scode = first_significant_code(code + 1+LINK_SIZE + xl, TRUE);
7210     register int op = *scode;     register int op = *scode;
7211    
7212     switch(op)     switch(op)
# Line 6713  do { Line 7215  do {
7215       return -1;       return -1;
7216    
7217       case OP_BRA:       case OP_BRA:
7218         case OP_BRAPOS:
7219       case OP_CBRA:       case OP_CBRA:
7220         case OP_SCBRA:
7221         case OP_CBRAPOS:
7222         case OP_SCBRAPOS:
7223       case OP_ASSERT:       case OP_ASSERT:
7224       case OP_ONCE:       case OP_ONCE:
7225         case OP_ONCE_NC:
7226       case OP_COND:       case OP_COND:
7227       if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)       if ((d = find_firstassertedchar(scode, op == OP_ASSERT)) < 0)
7228         return -1;         return -1;
7229       if (c < 0) c = d; else if (c != d) return -1;       if (c < 0) c = d; else if (c != d) return -1;
7230       break;       break;
7231    
7232       case OP_EXACT:       /* Fall through */       case OP_EXACT:
7233       scode += 2;       scode += 2;
7234         /* Fall through */
7235    
7236       case OP_CHAR:       case OP_CHAR:
      case OP_CHARNC:  
7237       case OP_PLUS:       case OP_PLUS:
7238       case OP_MINPLUS:       case OP_MINPLUS:
7239       case OP_POSPLUS:       case OP_POSPLUS:
7240       if (!inassert) return -1;       if (!inassert) return -1;
7241       if (c < 0)       if (c < 0) c = scode[1];
7242         {         else if (c != scode[1]) return -1;
7243         c = scode[1];       break;
7244         if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;  
7245         }       case OP_EXACTI:
7246       else if (c != scode[1]) return -1;       scode += 2;
7247         /* Fall through */
7248    
7249         case OP_CHARI:
7250         case OP_PLUSI:
7251         case OP_MINPLUSI:
7252         case OP_POSPLUSI:
7253         if (!inassert) return -1;
7254         if (c < 0) c = scode[1] | REQ_CASELESS;
7255           else if (c != scode[1]) return -1;
7256       break;       break;
7257       }       }
7258    
# Line 6799  compile_data *cd = &compile_block; Line 7315  compile_data *cd = &compile_block;
7315  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
7316  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
7317  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
7318  to fill in forward references to subpatterns. */  to fill in forward references to subpatterns. That may overflow, in which case
7319    new memory is obtained from malloc(). */
7320    
7321  uschar cworkspace[COMPILE_WORK_SIZE];  uschar cworkspace[COMPILE_WORK_SIZE];
7322    
# Line 6859  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7376  while (ptr[skipatstart] == CHAR_LEFT_PAR
7376      { skipatstart += 7; options |= PCRE_UTF8; continue; }      { skipatstart += 7; options |= PCRE_UTF8; continue; }
7377    else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)    else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
7378      { skipatstart += 6; options |= PCRE_UCP; continue; }      { skipatstart += 6; options |= PCRE_UCP; continue; }
7379      else if (strncmp((char *)(ptr+skipatstart+2), STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
7380        { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
7381    
7382    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
7383      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
# Line 6885  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7404  while (ptr[skipatstart] == CHAR_LEFT_PAR
7404    
7405  utf8 = (options & PCRE_UTF8) != 0;  utf8 = (options & PCRE_UTF8) != 0;
7406    
7407  /* 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
7408    return of an error code from _pcre_valid_utf8() is a new feature, introduced in
7409    release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
7410    not used here. */
7411    
7412  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
7413  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
7414       (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)       (errorcode = _pcre_valid_utf8((USPTR)pattern, -1, erroroffset)) != 0)
7415    {    {
7416    errorcode = ERR44;    errorcode = ERR44;
7417    goto PCRE_EARLY_ERROR_RETURN2;    goto PCRE_EARLY_ERROR_RETURN2;
# Line 6914  if ((options & PCRE_UCP) != 0) Line 7436  if ((options & PCRE_UCP) != 0)
7436    
7437  /* Check validity of \R options. */  /* Check validity of \R options. */
7438    
7439  switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))  if ((options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) ==
7440         (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
7441    {    {
7442    case 0:    errorcode = ERR56;
7443    case PCRE_BSR_ANYCRLF:    goto PCRE_EARLY_ERROR_RETURN;
   case PCRE_BSR_UNICODE:  
   break;  
   default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;  
7444    }    }
7445    
7446  /* 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 7506  cd->bracount = cd->final_bracount = 0;
7506  cd->names_found = 0;  cd->names_found = 0;
7507  cd->name_entry_size = 0;  cd->name_entry_size = 0;
7508  cd->name_table = NULL;  cd->name_table = NULL;
 cd->start_workspace = cworkspace;  
7509  cd->start_code = cworkspace;  cd->start_code = cworkspace;
7510  cd->hwm = cworkspace;  cd->hwm = cworkspace;
7511    cd->start_workspace = cworkspace;
7512    cd->workspace_size = COMPILE_WORK_SIZE;
7513  cd->start_pattern = (const uschar *)pattern;  cd->start_pattern = (const uschar *)pattern;
7514  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));  cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
7515  cd->req_varyopt = 0;  cd->req_varyopt = 0;
# Line 7005  outside can help speed up starting point Line 7526  outside can help speed up starting point
7526  ptr += skipatstart;  ptr += skipatstart;
7527  code = cworkspace;  code = cworkspace;
7528  *code = OP_BRA;  *code = OP_BRA;
7529  (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,  (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
7530    &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,    FALSE, 0, 0, &firstbyte, &reqbyte, NULL, cd, &length);
   &length);  
7531  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
7532    
7533  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
# Line 7024  externally provided function. Integer ov Line 7544  externally provided function. Integer ov
7544  because nowadays we limit the maximum value of cd->names_found and  because nowadays we limit the maximum value of cd->names_found and
7545  cd->name_entry_size. */  cd->name_entry_size. */
7546    
7547  size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);  size = length + sizeof(real_pcre) + cd->names_found * cd->name_entry_size;
7548  re = (real_pcre *)(pcre_malloc)(size);  re = (real_pcre *)(pcre_malloc)(size);
7549    
7550  if (re == NULL)  if (re == NULL)
# Line 7061  field; this time it's used for rememberi Line 7581  field; this time it's used for rememberi
7581  */  */
7582    
7583  cd->final_bracount = cd->bracount;  /* Save for checking forward references */  cd->final_bracount = cd->bracount;  /* Save for checking forward references */
7584    cd->assert_depth = 0;
7585  cd->bracount = 0;  cd->bracount = 0;
7586  cd->names_found = 0;  cd->names_found = 0;
7587  cd->name_table = (uschar *)re + re->name_table_offset;  cd->name_table = (uschar *)re + re->name_table_offset;
7588  codestart = cd->name_table + re->name_entry_size * re->name_count;  codestart = cd->name_table + re->name_entry_size * re->name_count;
7589  cd->start_code = codestart;  cd->start_code = codestart;
7590  cd->hwm = cworkspace;  cd->hwm = (uschar *)(cd->start_workspace);
7591  cd->req_varyopt = 0;  cd->req_varyopt = 0;
7592  cd->had_accept = FALSE;  cd->had_accept = FALSE;
7593  cd->check_lookbehind = FALSE;  cd->check_lookbehind = FALSE;
# Line 7079  of the function here. */ Line 7600  of the function here. */
7600  ptr = (const uschar *)pattern + skipatstart;  ptr = (const uschar *)pattern + skipatstart;
7601  code = (uschar *)codestart;  code = (uschar *)codestart;
7602  *code = OP_BRA;  *code = OP_BRA;
7603  (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,  (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0, 0,
7604    &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);    &firstbyte, &reqbyte, NULL, cd, NULL);
7605  re->top_bracket = cd->bracount;  re->top_bracket = cd->bracount;
7606  re->top_backref = cd->top_backref;  re->top_backref = cd->top_backref;
7607  re->flags = cd->external_flags;  re->flags = cd->external_flags;
7608    
7609  if (cd->had_accept) reqbyte = -1;   /* Must disable after (*ACCEPT) */  if (cd->had_accept) reqbyte = REQ_NONE;   /* Must disable after (*ACCEPT) */
7610    
7611  /* 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. */
7612    
# Line 7100  if debugging, leave the test till after Line 7621  if debugging, leave the test till after
7621  if (code - codestart > length) errorcode = ERR23;  if (code - codestart > length) errorcode = ERR23;
7622  #endif  #endif
7623    
7624  /* Fill in any forward references that are required. */  /* Fill in any forward references that are required. There may be repeated
7625    references; optimize for them, as searching a large regex takes time. */
7626    
7627  while (errorcode == 0 && cd->hwm > cworkspace)  if (cd->hwm > cd->start_workspace)
7628    {    {
7629    int offset, recno;    int prev_recno = -1;
7630    const uschar *groupptr;    const uschar *groupptr = NULL;
7631    cd->hwm -= LINK_SIZE;    while (errorcode == 0 && cd->hwm > cd->start_workspace)
7632    offset = GET(cd->hwm, 0);      {
7633    recno = GET(codestart, offset);      int offset, recno;
7634    groupptr = _pcre_find_bracket(codestart, utf8, recno);      cd->hwm -= LINK_SIZE;
7635    if (groupptr == NULL) errorcode = ERR53;      offset = GET(cd->hwm, 0);
7636      else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));      recno = GET(codestart, offset);
7637        if (recno != prev_recno)
7638          {
7639          groupptr = _pcre_find_bracket(codestart, utf8, recno);
7640          prev_recno = recno;
7641          }
7642        if (groupptr == NULL) errorcode = ERR53;
7643          else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
7644        }
7645    }    }
7646    
7647    /* If the workspace had to be expanded, free the new memory. */
7648    
7649    if (cd->workspace_size > COMPILE_WORK_SIZE)
7650      (pcre_free)((void *)cd->start_workspace);
7651    
7652  /* 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
7653  subpattern. */  subpattern. */
7654    
# Line 7146  if (cd->check_lookbehind) Line 7681  if (cd->check_lookbehind)
7681        uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);        uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
7682        int end_op = *be;        int end_op = *be;
7683        *be = OP_END;        *be = OP_END;
7684        fixed_length = find_fixedlength(cc, re->options, TRUE, cd);        fixed_length = find_fixedlength(cc, (re->options & PCRE_UTF8) != 0, TRUE,
7685            cd);
7686        *be = end_op;        *be = end_op;
7687        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
7688        if (fixed_length < 0)        if (fixed_length < 0)
7689          {          {
7690          errorcode = (fixed_length == -2)? ERR36 : ERR25;          errorcode = (fixed_length == -2)? ERR36 :
7691                        (fixed_length == -4)? ERR70 : ERR25;
7692          break;          break;
7693          }          }
7694        PUT(cc, 1, fixed_length);        PUT(cc, 1, fixed_length);
# Line 7185  start with ^. and also when all branches Line 7722  start with ^. and also when all branches
7722    
7723  if ((re->options & PCRE_ANCHORED) == 0)  if ((re->options & PCRE_ANCHORED) == 0)
7724    {    {
7725    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))  
7726      re->options |= PCRE_ANCHORED;      re->options |= PCRE_ANCHORED;
7727    else    else
7728      {      {
7729      if (firstbyte < 0)      if (firstbyte < 0)
7730        firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);        firstbyte = find_firstassertedchar(codestart, FALSE);
7731      if (firstbyte >= 0)   /* Remove caseless flag for non-caseable chars */      if (firstbyte >= 0)   /* Remove caseless flag for non-caseable chars */
7732        {        {
7733        int ch = firstbyte & 255;        int ch = firstbyte & 255;

Legend:
Removed from v.574  
changed lines
  Added in v.807

  ViewVC Help
Powered by ViewVC 1.1.5