/[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 406 by ph10, Mon Mar 23 12:05:43 2009 UTC revision 640 by ph10, Mon Jul 25 10:50:28 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) 1997-2009 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 53  supporting internal functions that are n Line 53  supporting internal functions that are n
53  #include "pcre_internal.h"  #include "pcre_internal.h"
54    
55    
56  /* When DEBUG is defined, we need the pcre_printint() function, which is also  /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57  used by pcretest. DEBUG is not defined when building a production library. */  also used by pcretest. PCRE_DEBUG is not defined when building a production
58    library. */
59    
60  #ifdef DEBUG  #ifdef PCRE_DEBUG
61  #include "pcre_printint.src"  #include "pcre_printint.src"
62  #endif  #endif
63    
# Line 91  is 4 there is plenty of room. */ Line 92  is 4 there is plenty of room. */
92    
93  #define COMPILE_WORK_SIZE (4096)  #define COMPILE_WORK_SIZE (4096)
94    
95    /* The overrun tests check for a slightly smaller size so that they detect the
96    overrun before it actually does run off the end of the data block. */
97    
98    #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99    
100    
101  /* Table for handling escaped characters in the range '0'-'z'. Positive returns  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102  are simple data values; negative values are for special things like \d and so  are simple data values; negative values are for special things like \d and so
# Line 118  static const short int escapes[] = { Line 124  static const short int escapes[] = {
124       -ESC_H,                  0,       -ESC_H,                  0,
125       0,                       -ESC_K,       0,                       -ESC_K,
126       0,                       0,       0,                       0,
127       0,                       0,       -ESC_N,                  0,
128       -ESC_P,                  -ESC_Q,       -ESC_P,                  -ESC_Q,
129       -ESC_R,                  -ESC_S,       -ESC_R,                  -ESC_S,
130       0,                       0,       0,                       0,
# Line 165  static const short int escapes[] = { Line 171  static const short int escapes[] = {
171  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
172  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
173  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
174  /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,  /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
175  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
176  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
177  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
# Line 182  string is built from string macros so th Line 188  string is built from string macros so th
188  platforms. */  platforms. */
189    
190  typedef struct verbitem {  typedef struct verbitem {
191    int   len;    int   len;                 /* Length of verb name */
192    int   op;    int   op;                  /* Op when no arg, or -1 if arg mandatory */
193      int   op_arg;              /* Op when arg present, or -1 if not allowed */
194  } verbitem;  } verbitem;
195    
196  static const char verbnames[] =  static const char verbnames[] =
197      "\0"                       /* Empty name is a shorthand for MARK */
198      STRING_MARK0
199    STRING_ACCEPT0    STRING_ACCEPT0
200    STRING_COMMIT0    STRING_COMMIT0
201    STRING_F0    STRING_F0
# Line 196  static const char verbnames[] = Line 205  static const char verbnames[] =
205    STRING_THEN;    STRING_THEN;
206    
207  static const verbitem verbs[] = {  static const verbitem verbs[] = {
208    { 6, OP_ACCEPT },    { 0, -1,        OP_MARK },
209    { 6, OP_COMMIT },    { 4, -1,        OP_MARK },
210    { 1, OP_FAIL },    { 6, OP_ACCEPT, -1 },
211    { 4, OP_FAIL },    { 6, OP_COMMIT, -1 },
212    { 5, OP_PRUNE },    { 1, OP_FAIL,   -1 },
213    { 4, OP_SKIP  },    { 4, OP_FAIL,   -1 },
214    { 4, OP_THEN  }    { 5, OP_PRUNE,  OP_PRUNE_ARG },
215      { 4, OP_SKIP,   OP_SKIP_ARG  },
216      { 4, OP_THEN,   OP_THEN_ARG  }
217  };  };
218    
219  static const int verbcount = sizeof(verbs)/sizeof(verbitem);  static const int verbcount = sizeof(verbs)/sizeof(verbitem);
# Line 250  static const int posix_class_maps[] = { Line 261  static const int posix_class_maps[] = {
261    cbit_xdigit,-1,          0              /* xdigit */    cbit_xdigit,-1,          0              /* xdigit */
262  };  };
263    
264    /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265    substitutes must be in the order of the names, defined above, and there are
266    both positive and negative cases. NULL means no substitute. */
267    
268    #ifdef SUPPORT_UCP
269    static const uschar *substitutes[] = {
270      (uschar *)"\\P{Nd}",    /* \D */
271      (uschar *)"\\p{Nd}",    /* \d */
272      (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
273      (uschar *)"\\p{Xsp}",   /* \s */
274      (uschar *)"\\P{Xwd}",   /* \W */
275      (uschar *)"\\p{Xwd}"    /* \w */
276    };
277    
278    static const uschar *posix_substitutes[] = {
279      (uschar *)"\\p{L}",     /* alpha */
280      (uschar *)"\\p{Ll}",    /* lower */
281      (uschar *)"\\p{Lu}",    /* upper */
282      (uschar *)"\\p{Xan}",   /* alnum */
283      NULL,                   /* ascii */
284      (uschar *)"\\h",        /* blank */
285      NULL,                   /* cntrl */
286      (uschar *)"\\p{Nd}",    /* digit */
287      NULL,                   /* graph */
288      NULL,                   /* print */
289      NULL,                   /* punct */
290      (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
291      (uschar *)"\\p{Xwd}",   /* word */
292      NULL,                   /* xdigit */
293      /* Negated cases */
294      (uschar *)"\\P{L}",     /* ^alpha */
295      (uschar *)"\\P{Ll}",    /* ^lower */
296      (uschar *)"\\P{Lu}",    /* ^upper */
297      (uschar *)"\\P{Xan}",   /* ^alnum */
298      NULL,                   /* ^ascii */
299      (uschar *)"\\H",        /* ^blank */
300      NULL,                   /* ^cntrl */
301      (uschar *)"\\P{Nd}",    /* ^digit */
302      NULL,                   /* ^graph */
303      NULL,                   /* ^print */
304      NULL,                   /* ^punct */
305      (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
306      (uschar *)"\\P{Xwd}",   /* ^word */
307      NULL                    /* ^xdigit */
308    };
309    #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310    #endif
311    
312  #define STRING(a)  # a  #define STRING(a)  # a
313  #define XSTRING(s) STRING(s)  #define XSTRING(s) STRING(s)
# Line 262  the number of relocations needed when a Line 320  the number of relocations needed when a
320  it is now one long string. We cannot use a table of offsets, because the  it is now one long string. We cannot use a table of offsets, because the
321  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322  simply count through to the one we want - this isn't a performance issue  simply count through to the one we want - this isn't a performance issue
323  because these strings are used only when there is a compilation error. */  because these strings are used only when there is a compilation error.
324    
325    Each substring ends with \0 to insert a null character. This includes the final
326    substring, so that the whole string ends with \0\0, which can be detected when
327    counting through. */
328    
329  static const char error_texts[] =  static const char error_texts[] =
330    "no error\0"    "no error\0"
# Line 309  static const char error_texts[] = Line 371  static const char error_texts[] =
371    /* 35 */    /* 35 */
372    "invalid condition (?(0)\0"    "invalid condition (?(0)\0"
373    "\\C not allowed in lookbehind assertion\0"    "\\C not allowed in lookbehind assertion\0"
374    "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"    "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375    "number after (?C is > 255\0"    "number after (?C is > 255\0"
376    "closing ) for (?C expected\0"    "closing ) for (?C expected\0"
377    /* 40 */    /* 40 */
# Line 331  static const char error_texts[] = Line 393  static const char error_texts[] =
393    "internal error: previously-checked referenced subpattern not found\0"    "internal error: previously-checked referenced subpattern not found\0"
394    "DEFINE group contains more than one branch\0"    "DEFINE group contains more than one branch\0"
395    /* 55 */    /* 55 */
396    "repeating a DEFINE group is not allowed\0"    "repeating a DEFINE group is not allowed\0"  /** DEAD **/
397    "inconsistent NEWLINE options\0"    "inconsistent NEWLINE options\0"
398    "\\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"
399    "a numbered reference must not be zero\0"    "a numbered reference must not be zero\0"
400    "(*VERB) with an argument is not supported\0"    "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401    /* 60 */    /* 60 */
402    "(*VERB) not recognized\0"    "(*VERB) not recognized\0"
403    "number is too big\0"    "number is too big\0"
404    "subpattern name expected\0"    "subpattern name expected\0"
405    "digit expected after (?+\0"    "digit expected after (?+\0"
406    "] is an invalid data character in JavaScript compatibility mode";    "] is an invalid data character in JavaScript compatibility mode\0"
407      /* 65 */
408      "different names for subpatterns of the same number are not allowed\0"
409      "(*MARK) must have an argument\0"
410      "this version of PCRE is not compiled with PCRE_UCP support\0"
411      "\\c must be followed by an ASCII character\0"
412      "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413      ;
414    
415  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
416  patterns. Note that the tables in chartables are dependent on the locale, and  patterns. Note that the tables in chartables are dependent on the locale, and
# Line 478  static const unsigned char ebcdic_charta Line 546  static const unsigned char ebcdic_charta
546  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
547    
548  static BOOL  static BOOL
549    compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
550      int *, int *, branch_chain *, compile_data *, int *);      int *, branch_chain *, compile_data *, int *);
551    
552    
553    
# Line 500  static const char * Line 568  static const char *
568  find_error_text(int n)  find_error_text(int n)
569  {  {
570  const char *s = error_texts;  const char *s = error_texts;
571  for (; n > 0; n--) while (*s++ != 0) {};  for (; n > 0; n--)
572      {
573      while (*s++ != 0) {};
574      if (*s == 0) return "Error text not found (please report)";
575      }
576  return s;  return s;
577  }  }
578    
579    
580  /*************************************************  /*************************************************
581    *            Check for counted repeat            *
582    *************************************************/
583    
584    /* This function is called when a '{' is encountered in a place where it might
585    start a quantifier. It looks ahead to see if it really is a quantifier or not.
586    It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
587    where the ddds are digits.
588    
589    Arguments:
590      p         pointer to the first char after '{'
591    
592    Returns:    TRUE or FALSE
593    */
594    
595    static BOOL
596    is_counted_repeat(const uschar *p)
597    {
598    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
599    while ((digitab[*p] & ctype_digit) != 0) p++;
600    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
601    
602    if (*p++ != CHAR_COMMA) return FALSE;
603    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604    
605    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
606    while ((digitab[*p] & ctype_digit) != 0) p++;
607    
608    return (*p == CHAR_RIGHT_CURLY_BRACKET);
609    }
610    
611    
612    
613    /*************************************************
614  *            Handle escapes                      *  *            Handle escapes                      *
615  *************************************************/  *************************************************/
616    
# Line 571  else Line 676  else
676    
677      case CHAR_l:      case CHAR_l:
678      case CHAR_L:      case CHAR_L:
     case CHAR_N:  
679      case CHAR_u:      case CHAR_u:
680      case CHAR_U:      case CHAR_U:
681      *errorcodeptr = ERR37;      *errorcodeptr = ERR37;
682      break;      break;
683    
684      /* \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
685        class, \g must be followed by one of a number of specific things:
686    
687      (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
688      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 594  else Line 699  else
699      the -ESC_g code (cf \k). */      the -ESC_g code (cf \k). */
700    
701      case CHAR_g:      case CHAR_g:
702        if (isclass) break;
703      if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)      if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
704        {        {
705        c = -ESC_g;        c = -ESC_g;
# Line 772  else Line 878  else
878      break;      break;
879    
880      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
881      This coding is ASCII-specific, but then the whole concept of \cx is      An error is given if the byte following \c is not an ASCII character. This
882        coding is ASCII-specific, but then the whole concept of \cx is
883      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
884    
885      case CHAR_c:      case CHAR_c:
# Line 782  else Line 889  else
889        *errorcodeptr = ERR2;        *errorcodeptr = ERR2;
890        break;        break;
891        }        }
892    #ifndef EBCDIC    /* ASCII/UTF-8 coding */
893  #ifndef EBCDIC  /* ASCII/UTF-8 coding */      if (c > 127)  /* Excludes all non-ASCII in either mode */
894          {
895          *errorcodeptr = ERR68;
896          break;
897          }
898      if (c >= CHAR_a && c <= CHAR_z) c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
899      c ^= 0x40;      c ^= 0x40;
900  #else           /* EBCDIC coding */  #else             /* EBCDIC coding */
901      if (c >= CHAR_a && c <= CHAR_z) c += 64;      if (c >= CHAR_a && c <= CHAR_z) c += 64;
902      c ^= 0xC0;      c ^= 0xC0;
903  #endif  #endif
# Line 809  else Line 920  else
920      }      }
921    }    }
922    
923    /* Perl supports \N{name} for character names, as well as plain \N for "not
924    newline". PCRE does not support \N{name}. However, it does support
925    quantification such as \N{2,3}. */
926    
927    if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
928         !is_counted_repeat(ptr+2))
929      *errorcodeptr = ERR37;
930    
931    /* If PCRE_UCP is set, we change the values for \d etc. */
932    
933    if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
934      c -= (ESC_DU - ESC_D);
935    
936    /* Set the pointer to the final character before returning. */
937    
938  *ptrptr = ptr;  *ptrptr = ptr;
939  return c;  return c;
940  }  }
# Line 909  return -1; Line 1035  return -1;
1035    
1036    
1037  /*************************************************  /*************************************************
 *            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);  
 }  
   
   
   
 /*************************************************  
1038  *         Read repeat counts                     *  *         Read repeat counts                     *
1039  *************************************************/  *************************************************/
1040    
# Line 1009  return p; Line 1102  return p;
1102    
1103    
1104  /*************************************************  /*************************************************
1105  *       Find forward referenced subpattern       *  *  Subroutine for finding forward reference      *
1106  *************************************************/  *************************************************/
1107    
1108  /* This function scans along a pattern's text looking for capturing  /* This recursive function is called only from find_parens() below. The
1109    top-level call starts at the beginning of the pattern. All other calls must
1110    start at a parenthesis. It scans along a pattern's text looking for capturing
1111  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
1112  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
1113  returns when it reaches a given numbered subpattern. This is used for forward  returns when it reaches a given numbered subpattern. Recursion is used to keep
1114  references to subpatterns. We know that if (?P< is encountered, the name will  track of subpatterns that reset the capturing group numbers - the (?| feature.
1115  be terminated by '>' because that is checked in the first pass.  
1116    This function was originally called only from the second pass, in which we know
1117    that if (?< or (?' or (?P< is encountered, the name will be correctly
1118    terminated because that is checked in the first pass. There is now one call to
1119    this function in the first pass, to check for a recursive back reference by
1120    name (so that we can make the whole group atomic). In this case, we need check
1121    only up to the current position in the pattern, and that is still OK because
1122    and previous occurrences will have been checked. To make this work, the test
1123    for "end of pattern" is a check against cd->end_pattern in the main loop,
1124    instead of looking for a binary zero. This means that the special first-pass
1125    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1126    processing items within the loop are OK, because afterwards the main loop will
1127    terminate.)
1128    
1129  Arguments:  Arguments:
1130    ptr          current position in the pattern    ptrptr       address of the current character pointer (updated)
1131    cd           compile background data    cd           compile background data
1132    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1133    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1134    xmode        TRUE if we are in /x mode    xmode        TRUE if we are in /x mode
1135      utf8         TRUE if we are in UTF-8 mode
1136      count        pointer to the current capturing subpattern number (updated)
1137    
1138  Returns:       the number of the named subpattern, or -1 if not found  Returns:       the number of the named subpattern, or -1 if not found
1139  */  */
1140    
1141  static int  static int
1142  find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,  find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1143    BOOL xmode)    BOOL xmode, BOOL utf8, int *count)
1144  {  {
1145  const uschar *thisname;  uschar *ptr = *ptrptr;
1146  int count = cd->bracount;  int start_count = *count;
1147    int hwm_count = start_count;
1148    BOOL dup_parens = FALSE;
1149    
1150  for (; *ptr != 0; ptr++)  /* If the first character is a parenthesis, check on the type of group we are
1151    dealing with. The very first call may not start with a parenthesis. */
1152    
1153    if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1154    {    {
1155    int term;    /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1156    
1157      if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1158    
1159      /* Handle a normal, unnamed capturing parenthesis. */
1160    
1161      else if (ptr[1] != CHAR_QUESTION_MARK)
1162        {
1163        *count += 1;
1164        if (name == NULL && *count == lorn) return *count;
1165        ptr++;
1166        }
1167    
1168      /* All cases now have (? at the start. Remember when we are in a group
1169      where the parenthesis numbers are duplicated. */
1170    
1171      else if (ptr[2] == CHAR_VERTICAL_LINE)
1172        {
1173        ptr += 3;
1174        dup_parens = TRUE;
1175        }
1176    
1177      /* Handle comments; all characters are allowed until a ket is reached. */
1178    
1179      else if (ptr[2] == CHAR_NUMBER_SIGN)
1180        {
1181        for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1182        goto FAIL_EXIT;
1183        }
1184    
1185      /* Handle a condition. If it is an assertion, just carry on so that it
1186      is processed as normal. If not, skip to the closing parenthesis of the
1187      condition (there can't be any nested parens). */
1188    
1189      else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1190        {
1191        ptr += 2;
1192        if (ptr[1] != CHAR_QUESTION_MARK)
1193          {
1194          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1195          if (*ptr != 0) ptr++;
1196          }
1197        }
1198    
1199      /* Start with (? but not a condition. */
1200    
1201      else
1202        {
1203        ptr += 2;
1204        if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
1205    
1206        /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1207    
1208        if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1209            ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1210          {
1211          int term;
1212          const uschar *thisname;
1213          *count += 1;
1214          if (name == NULL && *count == lorn) return *count;
1215          term = *ptr++;
1216          if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1217          thisname = ptr;
1218          while (*ptr != term) ptr++;
1219          if (name != NULL && lorn == ptr - thisname &&
1220              strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1221            return *count;
1222          term++;
1223          }
1224        }
1225      }
1226    
1227    /* Past any initial parenthesis handling, scan for parentheses or vertical
1228    bars. Stop if we get to cd->end_pattern. Note that this is important for the
1229    first-pass call when this value is temporarily adjusted to stop at the current
1230    position. So DO NOT change this to a test for binary zero. */
1231    
1232    for (; ptr < cd->end_pattern; ptr++)
1233      {
1234    /* Skip over backslashed characters and also entire \Q...\E */    /* Skip over backslashed characters and also entire \Q...\E */
1235    
1236    if (*ptr == CHAR_BACKSLASH)    if (*ptr == CHAR_BACKSLASH)
1237      {      {
1238      if (*(++ptr) == 0) return -1;      if (*(++ptr) == 0) goto FAIL_EXIT;
1239      if (*ptr == CHAR_Q) for (;;)      if (*ptr == CHAR_Q) for (;;)
1240        {        {
1241        while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};        while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1242        if (*ptr == 0) return -1;        if (*ptr == 0) goto FAIL_EXIT;
1243        if (*(++ptr) == CHAR_E) break;        if (*(++ptr) == CHAR_E) break;
1244        }        }
1245      continue;      continue;
# Line 1065  for (; *ptr != 0; ptr++) Line 1256  for (; *ptr != 0; ptr++)
1256      BOOL negate_class = FALSE;      BOOL negate_class = FALSE;
1257      for (;;)      for (;;)
1258        {        {
1259        int c = *(++ptr);        if (ptr[1] == CHAR_BACKSLASH)
       if (c == CHAR_BACKSLASH)  
1260          {          {
1261          if (ptr[1] == CHAR_E)          if (ptr[2] == CHAR_E)
1262            ptr++;            ptr+= 2;
1263          else if (strncmp((const char *)ptr+1,          else if (strncmp((const char *)ptr+2,
1264                   STR_Q STR_BACKSLASH STR_E, 3) == 0)                   STR_Q STR_BACKSLASH STR_E, 3) == 0)
1265            ptr += 3;            ptr += 4;
1266          else          else
1267            break;            break;
1268          }          }
1269        else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)        else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1270            {
1271          negate_class = TRUE;          negate_class = TRUE;
1272            ptr++;
1273            }
1274        else break;        else break;
1275        }        }
1276    
# Line 1093  for (; *ptr != 0; ptr++) Line 1286  for (; *ptr != 0; ptr++)
1286        if (*ptr == 0) return -1;        if (*ptr == 0) return -1;
1287        if (*ptr == CHAR_BACKSLASH)        if (*ptr == CHAR_BACKSLASH)
1288          {          {
1289          if (*(++ptr) == 0) return -1;          if (*(++ptr) == 0) goto FAIL_EXIT;
1290          if (*ptr == CHAR_Q) for (;;)          if (*ptr == CHAR_Q) for (;;)
1291            {            {
1292            while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};            while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1293            if (*ptr == 0) return -1;            if (*ptr == 0) goto FAIL_EXIT;
1294            if (*(++ptr) == CHAR_E) break;            if (*(++ptr) == CHAR_E) break;
1295            }            }
1296          continue;          continue;
# Line 1110  for (; *ptr != 0; ptr++) Line 1303  for (; *ptr != 0; ptr++)
1303    
1304    if (xmode && *ptr == CHAR_NUMBER_SIGN)    if (xmode && *ptr == CHAR_NUMBER_SIGN)
1305      {      {
1306      while (*(++ptr) != 0 && *ptr != CHAR_NL) {};      ptr++;
1307      if (*ptr == 0) return -1;      while (*ptr != 0)
1308          {
1309          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1310          ptr++;
1311    #ifdef SUPPORT_UTF8
1312          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1313    #endif
1314          }
1315        if (*ptr == 0) goto FAIL_EXIT;
1316      continue;      continue;
1317      }      }
1318    
1319    /* An opening parens must now be a real metacharacter */    /* Check for the special metacharacters */
1320    
1321    if (*ptr != CHAR_LEFT_PARENTHESIS) continue;    if (*ptr == CHAR_LEFT_PARENTHESIS)
   if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)  
1322      {      {
1323      count++;      int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1324      if (name == NULL && count == lorn) return count;      if (rc > 0) return rc;
1325      continue;      if (*ptr == 0) goto FAIL_EXIT;
1326      }      }
1327    
1328    ptr += 2;    else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1329    if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */      {
1330        if (dup_parens && *count < hwm_count) *count = hwm_count;
1331        goto FAIL_EXIT;
1332        }
1333    
1334    /* We have to disambiguate (?<! and (?<= from (?<name> */    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1335        {
1336        if (*count > hwm_count) hwm_count = *count;
1337        *count = start_count;
1338        }
1339      }
1340    
1341    FAIL_EXIT:
1342    *ptrptr = ptr;
1343    return -1;
1344    }
1345    
   if ((*ptr != CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_EXCLAMATION_MARK ||  
       ptr[1] == CHAR_EQUALS_SIGN) && *ptr != CHAR_APOSTROPHE)  
     continue;  
1346    
   count++;  
1347    
1348    if (name == NULL && count == lorn) return count;  
1349    term = *ptr++;  /*************************************************
1350    if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;  *       Find forward referenced subpattern       *
1351    thisname = ptr;  *************************************************/
1352    while (*ptr != term) ptr++;  
1353    if (name != NULL && lorn == ptr - thisname &&  /* This function scans along a pattern's text looking for capturing
1354        strncmp((const char *)name, (const char *)thisname, lorn) == 0)  subpatterns, and counting them. If it finds a named pattern that matches the
1355      return count;  name it is given, it returns its number. Alternatively, if the name is NULL, it
1356    returns when it reaches a given numbered subpattern. This is used for forward
1357    references to subpatterns. We used to be able to start this scan from the
1358    current compiling point, using the current count value from cd->bracount, and
1359    do it all in a single loop, but the addition of the possibility of duplicate
1360    subpattern numbers means that we have to scan from the very start, in order to
1361    take account of such duplicates, and to use a recursive function to keep track
1362    of the different types of group.
1363    
1364    Arguments:
1365      cd           compile background data
1366      name         name to seek, or NULL if seeking a numbered subpattern
1367      lorn         name length, or subpattern number if name is NULL
1368      xmode        TRUE if we are in /x mode
1369      utf8         TRUE if we are in UTF-8 mode
1370    
1371    Returns:       the number of the found subpattern, or -1 if not found
1372    */
1373    
1374    static int
1375    find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1376      BOOL utf8)
1377    {
1378    uschar *ptr = (uschar *)cd->start_pattern;
1379    int count = 0;
1380    int rc;
1381    
1382    /* If the pattern does not start with an opening parenthesis, the first call
1383    to find_parens_sub() will scan right to the end (if necessary). However, if it
1384    does start with a parenthesis, find_parens_sub() will return when it hits the
1385    matching closing parens. That is why we have to have a loop. */
1386    
1387    for (;;)
1388      {
1389      rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1390      if (rc > 0 || *ptr++ == 0) break;
1391    }    }
1392    
1393  return -1;  return rc;
1394  }  }
1395    
1396    
1397    
1398    
1399  /*************************************************  /*************************************************
1400  *      Find first significant op code            *  *      Find first significant op code            *
1401  *************************************************/  *************************************************/
1402    
1403  /* This is called by several functions that scan a compiled expression looking  /* This is called by several functions that scan a compiled expression looking
1404  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
1405  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
1406  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
1407  assertions, and also the \b assertion; for others it does not.  does not.
1408    
1409  Arguments:  Arguments:
1410    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  
1411    skipassert   TRUE if certain assertions are to be skipped    skipassert   TRUE if certain assertions are to be skipped
1412    
1413  Returns:       pointer to the first significant opcode  Returns:       pointer to the first significant opcode
1414  */  */
1415    
1416  static const uschar*  static const uschar*
1417  first_significant_code(const uschar *code, int *options, int optbit,  first_significant_code(const uschar *code, BOOL skipassert)
   BOOL skipassert)  
1418  {  {
1419  for (;;)  for (;;)
1420    {    {
1421    switch ((int)*code)    switch ((int)*code)
1422      {      {
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
   
1423      case OP_ASSERT_NOT:      case OP_ASSERT_NOT:
1424      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1425      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
# Line 1200  for (;;) Line 1435  for (;;)
1435    
1436      case OP_CALLOUT:      case OP_CALLOUT:
1437      case OP_CREF:      case OP_CREF:
1438        case OP_NCREF:
1439      case OP_RREF:      case OP_RREF:
1440        case OP_NRREF:
1441      case OP_DEF:      case OP_DEF:
1442      code += _pcre_OP_lengths[*code];      code += _pcre_OP_lengths[*code];
1443      break;      break;
# Line 1216  for (;;) Line 1453  for (;;)
1453    
1454    
1455  /*************************************************  /*************************************************
1456  *        Find the fixed length of a pattern      *  *        Find the fixed length of a branch       *
1457  *************************************************/  *************************************************/
1458    
1459  /* Scan a pattern and compute the fixed length of subject that will match it,  /* Scan a branch and compute the fixed length of subject that will match it,
1460  if the length is fixed. This is needed for dealing with backward assertions.  if the length is fixed. This is needed for dealing with backward assertions.
1461  In UTF8 mode, the result is in characters rather than bytes.  In UTF8 mode, the result is in characters rather than bytes. The branch is
1462    temporarily terminated with OP_END when this function is called.
1463    
1464    This function is called when a backward assertion is encountered, so that if it
1465    fails, the error message can point to the correct place in the pattern.
1466    However, we cannot do this when the assertion contains subroutine calls,
1467    because they can be forward references. We solve this by remembering this case
1468    and doing the check at the end; a flag specifies which mode we are running in.
1469    
1470  Arguments:  Arguments:
1471    code     points to the start of the pattern (the bracket)    code     points to the start of the pattern (the bracket)
1472    options  the compiling options    utf8     TRUE in UTF-8 mode
1473      atend    TRUE if called when the pattern is complete
1474      cd       the "compile data" structure
1475    
1476  Returns:   the fixed length, or -1 if there is no fixed length,  Returns:   the fixed length,
1477                 or -1 if there is no fixed length,
1478               or -2 if \C was encountered               or -2 if \C was encountered
1479                 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1480  */  */
1481    
1482  static int  static int
1483  find_fixedlength(uschar *code, int options)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1484  {  {
1485  int length = -1;  int length = -1;
1486    
# Line 1245  branch, check the length against that of Line 1493  branch, check the length against that of
1493  for (;;)  for (;;)
1494    {    {
1495    int d;    int d;
1496      uschar *ce, *cs;
1497    register int op = *cc;    register int op = *cc;
1498    switch (op)    switch (op)
1499      {      {
1500        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1501        OP_BRA (normal non-capturing bracket) because the other variants of these
1502        opcodes are all concerned with unlimited repeated groups, which of course
1503        are not of fixed length. They will cause a -1 response from the default
1504        case of this switch. */
1505    
1506      case OP_CBRA:      case OP_CBRA:
1507      case OP_BRA:      case OP_BRA:
1508      case OP_ONCE:      case OP_ONCE:
1509      case OP_COND:      case OP_COND:
1510      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1511      if (d < 0) return d;      if (d < 0) return d;
1512      branchlength += d;      branchlength += d;
1513      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
# Line 1261  for (;;) Line 1516  for (;;)
1516    
1517      /* 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
1518      call. If it's ALT it is an alternation in a nested call. If it is      call. If it's ALT it is an alternation in a nested call. If it is
1519      END it's the end of the outer call. All can be handled by the same code. */      END it's the end of the outer call. All can be handled by the same code.
1520        Note that we must not include the OP_KETRxxx opcodes here, because they
1521        all imply an unlimited repeat. */
1522    
1523      case OP_ALT:      case OP_ALT:
1524      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1525      case OP_END:      case OP_END:
1526      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1527        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
# Line 1275  for (;;) Line 1530  for (;;)
1530      branchlength = 0;      branchlength = 0;
1531      break;      break;
1532    
1533        /* A true recursion implies not fixed length, but a subroutine call may
1534        be OK. If the subroutine is a forward reference, we can't deal with
1535        it until the end of the pattern, so return -3. */
1536    
1537        case OP_RECURSE:
1538        if (!atend) return -3;
1539        cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1540        do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1541        if (cc > cs && cc < ce) return -1;                /* Recursion */
1542        d = find_fixedlength(cs + 2, utf8, atend, cd);
1543        if (d < 0) return d;
1544        branchlength += d;
1545        cc += 1 + LINK_SIZE;
1546        break;
1547    
1548      /* Skip over assertive subpatterns */      /* Skip over assertive subpatterns */
1549    
1550      case OP_ASSERT:      case OP_ASSERT:
# Line 1288  for (;;) Line 1558  for (;;)
1558    
1559      case OP_REVERSE:      case OP_REVERSE:
1560      case OP_CREF:      case OP_CREF:
1561        case OP_NCREF:
1562      case OP_RREF:      case OP_RREF:
1563        case OP_NRREF:
1564      case OP_DEF:      case OP_DEF:
     case OP_OPT:  
1565      case OP_CALLOUT:      case OP_CALLOUT:
1566      case OP_SOD:      case OP_SOD:
1567      case OP_SOM:      case OP_SOM:
1568        case OP_SET_SOM:
1569      case OP_EOD:      case OP_EOD:
1570      case OP_EODN:      case OP_EODN:
1571      case OP_CIRC:      case OP_CIRC:
1572        case OP_CIRCM:
1573      case OP_DOLL:      case OP_DOLL:
1574        case OP_DOLLM:
1575      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1576      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1577      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
# Line 1306  for (;;) Line 1580  for (;;)
1580      /* Handle literal characters */      /* Handle literal characters */
1581    
1582      case OP_CHAR:      case OP_CHAR:
1583      case OP_CHARNC:      case OP_CHARI:
1584      case OP_NOT:      case OP_NOT:
1585        case OP_NOTI:
1586      branchlength++;      branchlength++;
1587      cc += 2;      cc += 2;
1588  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1589      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while ((*cc & 0xc0) == 0x80) cc++;  
       }  
1590  #endif  #endif
1591      break;      break;
1592    
# Line 1325  for (;;) Line 1597  for (;;)
1597      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1598      cc += 4;      cc += 4;
1599  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1600      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while((*cc & 0x80) == 0x80) cc++;  
       }  
1601  #endif  #endif
1602      break;      break;
1603    
# Line 1407  for (;;) Line 1676  for (;;)
1676    
1677    
1678  /*************************************************  /*************************************************
1679  *    Scan compiled regex for numbered bracket    *  *    Scan compiled regex for specific bracket    *
1680  *************************************************/  *************************************************/
1681    
1682  /* This little function scans through a compiled pattern until it finds a  /* This little function scans through a compiled pattern until it finds a
1683  capturing bracket with the given number.  capturing bracket with the given number, or, if the number is negative, an
1684    instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1685    so that it can be called from pcre_study() when finding the minimum matching
1686    length.
1687    
1688  Arguments:  Arguments:
1689    code        points to start of expression    code        points to start of expression
1690    utf8        TRUE in UTF-8 mode    utf8        TRUE in UTF-8 mode
1691    number      the required bracket number    number      the required bracket number or negative to find a lookbehind
1692    
1693  Returns:      pointer to the opcode for the bracket, or NULL if not found  Returns:      pointer to the opcode for the bracket, or NULL if not found
1694  */  */
1695    
1696  static const uschar *  const uschar *
1697  find_bracket(const uschar *code, BOOL utf8, int number)  _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1698  {  {
1699  for (;;)  for (;;)
1700    {    {
1701    register int c = *code;    register int c = *code;
1702    
1703    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1704    
1705    /* 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 1435  for (;;) Line 1708  for (;;)
1708    
1709    if (c == OP_XCLASS) code += GET(code, 1);    if (c == OP_XCLASS) code += GET(code, 1);
1710    
1711      /* Handle recursion */
1712    
1713      else if (c == OP_REVERSE)
1714        {
1715        if (number < 0) return (uschar *)code;
1716        code += _pcre_OP_lengths[c];
1717        }
1718    
1719    /* Handle capturing bracket */    /* Handle capturing bracket */
1720    
1721    else if (c == OP_CBRA)    else if (c == OP_CBRA || c == OP_SCBRA ||
1722               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1723      {      {
1724      int n = GET2(code, 1+LINK_SIZE);      int n = GET2(code, 1+LINK_SIZE);
1725      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
# Line 1446  for (;;) Line 1728  for (;;)
1728    
1729    /* Otherwise, we can get the item's length from the table, except that for    /* Otherwise, we can get the item's length from the table, except that for
1730    repeated character types, we have to test for \p and \P, which have an extra    repeated character types, we have to test for \p and \P, which have an extra
1731    two bytes of parameters. */    two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1732      must add in its length. */
1733    
1734    else    else
1735      {      {
# Line 1470  for (;;) Line 1753  for (;;)
1753        case OP_TYPEPOSUPTO:        case OP_TYPEPOSUPTO:
1754        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1755        break;        break;
1756    
1757          case OP_MARK:
1758          case OP_PRUNE_ARG:
1759          case OP_SKIP_ARG:
1760          code += code[1];
1761          break;
1762    
1763          case OP_THEN_ARG:
1764          code += code[1+LINK_SIZE];
1765          break;
1766        }        }
1767    
1768      /* Add in the fixed length from the table */      /* Add in the fixed length from the table */
# Line 1484  for (;;) Line 1777  for (;;)
1777      if (utf8) switch(c)      if (utf8) switch(c)
1778        {        {
1779        case OP_CHAR:        case OP_CHAR:
1780        case OP_CHARNC:        case OP_CHARI:
1781        case OP_EXACT:        case OP_EXACT:
1782          case OP_EXACTI:
1783        case OP_UPTO:        case OP_UPTO:
1784          case OP_UPTOI:
1785        case OP_MINUPTO:        case OP_MINUPTO:
1786          case OP_MINUPTOI:
1787        case OP_POSUPTO:        case OP_POSUPTO:
1788          case OP_POSUPTOI:
1789        case OP_STAR:        case OP_STAR:
1790          case OP_STARI:
1791        case OP_MINSTAR:        case OP_MINSTAR:
1792          case OP_MINSTARI:
1793        case OP_POSSTAR:        case OP_POSSTAR:
1794          case OP_POSSTARI:
1795        case OP_PLUS:        case OP_PLUS:
1796          case OP_PLUSI:
1797        case OP_MINPLUS:        case OP_MINPLUS:
1798          case OP_MINPLUSI:
1799        case OP_POSPLUS:        case OP_POSPLUS:
1800          case OP_POSPLUSI:
1801        case OP_QUERY:        case OP_QUERY:
1802          case OP_QUERYI:
1803        case OP_MINQUERY:        case OP_MINQUERY:
1804          case OP_MINQUERYI:
1805        case OP_POSQUERY:        case OP_POSQUERY:
1806          case OP_POSQUERYI:
1807        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1808        break;        break;
1809        }        }
# Line 1541  for (;;) Line 1847  for (;;)
1847    
1848    /* Otherwise, we can get the item's length from the table, except that for    /* Otherwise, we can get the item's length from the table, except that for
1849    repeated character types, we have to test for \p and \P, which have an extra    repeated character types, we have to test for \p and \P, which have an extra
1850    two bytes of parameters. */    two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1851      must add in its length. */
1852    
1853    else    else
1854      {      {
# Line 1565  for (;;) Line 1872  for (;;)
1872        case OP_TYPEEXACT:        case OP_TYPEEXACT:
1873        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1874        break;        break;
1875    
1876          case OP_MARK:
1877          case OP_PRUNE_ARG:
1878          case OP_SKIP_ARG:
1879          code += code[1];
1880          break;
1881    
1882          case OP_THEN_ARG:
1883          code += code[1+LINK_SIZE];
1884          break;
1885        }        }
1886    
1887      /* Add in the fixed length from the table */      /* Add in the fixed length from the table */
# Line 1579  for (;;) Line 1896  for (;;)
1896      if (utf8) switch(c)      if (utf8) switch(c)
1897        {        {
1898        case OP_CHAR:        case OP_CHAR:
1899        case OP_CHARNC:        case OP_CHARI:
1900        case OP_EXACT:        case OP_EXACT:
1901          case OP_EXACTI:
1902        case OP_UPTO:        case OP_UPTO:
1903          case OP_UPTOI:
1904        case OP_MINUPTO:        case OP_MINUPTO:
1905          case OP_MINUPTOI:
1906        case OP_POSUPTO:        case OP_POSUPTO:
1907          case OP_POSUPTOI:
1908        case OP_STAR:        case OP_STAR:
1909          case OP_STARI:
1910        case OP_MINSTAR:        case OP_MINSTAR:
1911          case OP_MINSTARI:
1912        case OP_POSSTAR:        case OP_POSSTAR:
1913          case OP_POSSTARI:
1914        case OP_PLUS:        case OP_PLUS:
1915          case OP_PLUSI:
1916        case OP_MINPLUS:        case OP_MINPLUS:
1917          case OP_MINPLUSI:
1918        case OP_POSPLUS:        case OP_POSPLUS:
1919          case OP_POSPLUSI:
1920        case OP_QUERY:        case OP_QUERY:
1921          case OP_QUERYI:
1922        case OP_MINQUERY:        case OP_MINQUERY:
1923          case OP_MINQUERYI:
1924        case OP_POSQUERY:        case OP_POSQUERY:
1925          case OP_POSQUERYI:
1926        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1927        break;        break;
1928        }        }
# Line 1621  Arguments: Line 1951  Arguments:
1951    code        points to start of search    code        points to start of search
1952    endcode     points to where to stop    endcode     points to where to stop
1953    utf8        TRUE if in UTF8 mode    utf8        TRUE if in UTF8 mode
1954      cd          contains pointers to tables etc.
1955    
1956  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
1957  */  */
1958    
1959  static BOOL  static BOOL
1960  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1961      compile_data *cd)
1962  {  {
1963  register int c;  register int c;
1964  for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1965       code < endcode;       code < endcode;
1966       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1967    {    {
1968    const uschar *ccode;    const uschar *ccode;
1969    
# Line 1647  for (code = first_significant_code(code Line 1979  for (code = first_significant_code(code
1979      continue;      continue;
1980      }      }
1981    
1982      /* For a recursion/subroutine call, if its end has been reached, which
1983      implies a backward reference subroutine call, we can scan it. If it's a
1984      forward reference subroutine call, we can't. To detect forward reference
1985      we have to scan up the list that is kept in the workspace. This function is
1986      called only when doing the real compile, not during the pre-compile that
1987      measures the size of the compiled pattern. */
1988    
1989      if (c == OP_RECURSE)
1990        {
1991        const uschar *scode;
1992        BOOL empty_branch;
1993    
1994        /* Test for forward reference */
1995    
1996        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1997          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1998    
1999        /* Not a forward reference, test for completed backward reference */
2000    
2001        empty_branch = FALSE;
2002        scode = cd->start_code + GET(code, 1);
2003        if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2004    
2005        /* Completed backwards reference */
2006    
2007        do
2008          {
2009          if (could_be_empty_branch(scode, endcode, utf8, cd))
2010            {
2011            empty_branch = TRUE;
2012            break;
2013            }
2014          scode += GET(scode, 1);
2015          }
2016        while (*scode == OP_ALT);
2017    
2018        if (!empty_branch) return FALSE;  /* All branches are non-empty */
2019        continue;
2020        }
2021    
2022    /* Groups with zero repeats can of course be empty; skip them. */    /* Groups with zero repeats can of course be empty; skip them. */
2023    
2024    if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)    if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2025          c == OP_BRAPOSZERO)
2026      {      {
2027      code += _pcre_OP_lengths[c];      code += _pcre_OP_lengths[c];
2028      do code += GET(code, 1); while (*code == OP_ALT);      do code += GET(code, 1); while (*code == OP_ALT);
# Line 1657  for (code = first_significant_code(code Line 2030  for (code = first_significant_code(code
2030      continue;      continue;
2031      }      }
2032    
2033      /* A nested group that is already marked as "could be empty" can just be
2034      skipped. */
2035    
2036      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2037          c == OP_SCBRA || c == OP_SCBRAPOS)
2038        {
2039        do code += GET(code, 1); while (*code == OP_ALT);
2040        c = *code;
2041        continue;
2042        }
2043    
2044    /* For other groups, scan the branches. */    /* For other groups, scan the branches. */
2045    
2046    if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)    if (c == OP_BRA  || c == OP_BRAPOS ||
2047          c == OP_CBRA || c == OP_CBRAPOS ||
2048          c == OP_ONCE || c == OP_COND)
2049      {      {
2050      BOOL empty_branch;      BOOL empty_branch;
2051      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
# Line 1675  for (code = first_significant_code(code Line 2061  for (code = first_significant_code(code
2061        empty_branch = FALSE;        empty_branch = FALSE;
2062        do        do
2063          {          {
2064          if (!empty_branch && could_be_empty_branch(code, endcode, utf8))          if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2065            empty_branch = TRUE;            empty_branch = TRUE;
2066          code += GET(code, 1);          code += GET(code, 1);
2067          }          }
# Line 1746  for (code = first_significant_code(code Line 2132  for (code = first_significant_code(code
2132      case OP_ALLANY:      case OP_ALLANY:
2133      case OP_ANYBYTE:      case OP_ANYBYTE:
2134      case OP_CHAR:      case OP_CHAR:
2135      case OP_CHARNC:      case OP_CHARI:
2136      case OP_NOT:      case OP_NOT:
2137        case OP_NOTI:
2138      case OP_PLUS:      case OP_PLUS:
2139      case OP_MINPLUS:      case OP_MINPLUS:
2140      case OP_POSPLUS:      case OP_POSPLUS:
# Line 1787  for (code = first_significant_code(code Line 2174  for (code = first_significant_code(code
2174      case OP_KET:      case OP_KET:
2175      case OP_KETRMAX:      case OP_KETRMAX:
2176      case OP_KETRMIN:      case OP_KETRMIN:
2177        case OP_KETRPOS:
2178      case OP_ALT:      case OP_ALT:
2179      return TRUE;      return TRUE;
2180    
# Line 1795  for (code = first_significant_code(code Line 2183  for (code = first_significant_code(code
2183    
2184  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2185      case OP_STAR:      case OP_STAR:
2186        case OP_STARI:
2187      case OP_MINSTAR:      case OP_MINSTAR:
2188        case OP_MINSTARI:
2189      case OP_POSSTAR:      case OP_POSSTAR:
2190        case OP_POSSTARI:
2191      case OP_QUERY:      case OP_QUERY:
2192        case OP_QUERYI:
2193      case OP_MINQUERY:      case OP_MINQUERY:
2194        case OP_MINQUERYI:
2195      case OP_POSQUERY:      case OP_POSQUERY:
2196        case OP_POSQUERYI:
2197        if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2198        break;
2199    
2200      case OP_UPTO:      case OP_UPTO:
2201        case OP_UPTOI:
2202      case OP_MINUPTO:      case OP_MINUPTO:
2203        case OP_MINUPTOI:
2204      case OP_POSUPTO:      case OP_POSUPTO:
2205      if (utf8) while ((code[2] & 0xc0) == 0x80) code++;      case OP_POSUPTOI:
2206        if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2207      break;      break;
2208  #endif  #endif
2209    
2210        /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2211        string. */
2212    
2213        case OP_MARK:
2214        case OP_PRUNE_ARG:
2215        case OP_SKIP_ARG:
2216        code += code[1];
2217        break;
2218    
2219        case OP_THEN_ARG:
2220        code += code[1+LINK_SIZE];
2221        break;
2222    
2223        /* None of the remaining opcodes are required to match a character. */
2224    
2225        default:
2226        break;
2227      }      }
2228    }    }
2229    
# Line 1822  return TRUE; Line 2240  return TRUE;
2240  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
2241  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,
2242  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.
2243    This function is called only during the real compile, not during the
2244    pre-compile.
2245    
2246  Arguments:  Arguments:
2247    code        points to start of the recursion    code        points to start of the recursion
2248    endcode     points to where to stop (current RECURSE item)    endcode     points to where to stop (current RECURSE item)
2249    bcptr       points to the chain of current (unclosed) branch starts    bcptr       points to the chain of current (unclosed) branch starts
2250    utf8        TRUE if in UTF-8 mode    utf8        TRUE if in UTF-8 mode
2251      cd          pointers to tables etc
2252    
2253  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2254  */  */
2255    
2256  static BOOL  static BOOL
2257  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2258    BOOL utf8)    BOOL utf8, compile_data *cd)
2259  {  {
2260  while (bcptr != NULL && bcptr->current >= code)  while (bcptr != NULL && bcptr->current_branch >= code)
2261    {    {
2262    if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;    if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2263        return FALSE;
2264    bcptr = bcptr->outer;    bcptr = bcptr->outer;
2265    }    }
2266  return TRUE;  return TRUE;
# Line 1870  where Perl recognizes it as the POSIX cl Line 2292  where Perl recognizes it as the POSIX cl
2292  "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,
2293  I think.  I think.
2294    
2295    A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2296    It seems that the appearance of a nested POSIX class supersedes an apparent
2297    external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2298    a digit. Also, unescaped square brackets may also appear as part of class
2299    names. For example, [:a[:abc]b:] gives unknown class "[:abc]b:]"in Perl.
2300    
2301  Arguments:  Arguments:
2302    ptr      pointer to the initial [    ptr      pointer to the initial [
2303    endptr   where to return the end pointer    endptr   where to return the end pointer
# Line 1884  int terminator;          /* Don't combin Line 2312  int terminator;          /* Don't combin
2312  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2313  for (++ptr; *ptr != 0; ptr++)  for (++ptr; *ptr != 0; ptr++)
2314    {    {
2315    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2316        ptr++;
2317      else
2318      {      {
     if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;  
2319      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)      if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2320        {        {
2321        *endptr = ptr;        *endptr = ptr;
2322        return TRUE;        return TRUE;
2323        }        }
2324        if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2325             (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2326              ptr[1] == CHAR_EQUALS_SIGN) &&
2327            check_posix_syntax(ptr, endptr))
2328          return FALSE;
2329      }      }
2330    }    }
2331  return FALSE;  return FALSE;
# Line 2020  auto_callout(uschar *code, const uschar Line 2454  auto_callout(uschar *code, const uschar
2454  {  {
2455  *code++ = OP_CALLOUT;  *code++ = OP_CALLOUT;
2456  *code++ = 255;  *code++ = 255;
2457  PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
2458  PUT(code, LINK_SIZE, 0);                /* Default length */  PUT(code, LINK_SIZE, 0);                       /* Default length */
2459  return code + 2*LINK_SIZE;  return code + 2*LINK_SIZE;
2460  }  }
2461    
# Line 2046  Returns:             nothing Line 2480  Returns:             nothing
2480  static void  static void
2481  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2482  {  {
2483  int length = ptr - cd->start_pattern - GET(previous_callout, 2);  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2484  PUT(previous_callout, 2 + LINK_SIZE, length);  PUT(previous_callout, 2 + LINK_SIZE, length);
2485  }  }
2486    
# Line 2096  for (++c; c <= d; c++) Line 2530  for (++c; c <= d; c++)
2530    
2531  return TRUE;  return TRUE;
2532  }  }
2533    
2534    
2535    
2536    /*************************************************
2537    *        Check a character and a property        *
2538    *************************************************/
2539    
2540    /* This function is called by check_auto_possessive() when a property item
2541    is adjacent to a fixed character.
2542    
2543    Arguments:
2544      c            the character
2545      ptype        the property type
2546      pdata        the data for the type
2547      negated      TRUE if it's a negated property (\P or \p{^)
2548    
2549    Returns:       TRUE if auto-possessifying is OK
2550    */
2551    
2552    static BOOL
2553    check_char_prop(int c, int ptype, int pdata, BOOL negated)
2554    {
2555    const ucd_record *prop = GET_UCD(c);
2556    switch(ptype)
2557      {
2558      case PT_LAMP:
2559      return (prop->chartype == ucp_Lu ||
2560              prop->chartype == ucp_Ll ||
2561              prop->chartype == ucp_Lt) == negated;
2562    
2563      case PT_GC:
2564      return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2565    
2566      case PT_PC:
2567      return (pdata == prop->chartype) == negated;
2568    
2569      case PT_SC:
2570      return (pdata == prop->script) == negated;
2571    
2572      /* These are specials */
2573    
2574      case PT_ALNUM:
2575      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2576              _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2577    
2578      case PT_SPACE:    /* Perl space */
2579      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2580              c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2581              == negated;
2582    
2583      case PT_PXSPACE:  /* POSIX space */
2584      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2585              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2586              c == CHAR_FF || c == CHAR_CR)
2587              == negated;
2588    
2589      case PT_WORD:
2590      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2591              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2592              c == CHAR_UNDERSCORE) == negated;
2593      }
2594    return FALSE;
2595    }
2596  #endif  /* SUPPORT_UCP */  #endif  /* SUPPORT_UCP */
2597    
2598    
# Line 2109  whether the next thing could possibly ma Line 2606  whether the next thing could possibly ma
2606  sense to automatically possessify the repeated item.  sense to automatically possessify the repeated item.
2607    
2608  Arguments:  Arguments:
2609    op_code       the repeated op code    previous      pointer to the repeated opcode
   this          data for this item, depends on the opcode  
2610    utf8          TRUE in UTF-8 mode    utf8          TRUE in UTF-8 mode
   utf8_char     used for utf8 character bytes, NULL if not relevant  
2611    ptr           next character in pattern    ptr           next character in pattern
2612    options       options bits    options       options bits
2613    cd            contains pointers to tables etc.    cd            contains pointers to tables etc.
# Line 2121  Returns:        TRUE if possessifying is Line 2616  Returns:        TRUE if possessifying is
2616  */  */
2617    
2618  static BOOL  static BOOL
2619  check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,  check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2620    const uschar *ptr, int options, compile_data *cd)    int options, compile_data *cd)
2621  {  {
2622  int next;  int c, next;
2623    int op_code = *previous++;
2624    
2625  /* Skip whitespace and comments in extended mode */  /* Skip whitespace and comments in extended mode */
2626    
# Line 2135  if ((options & PCRE_EXTENDED) != 0) Line 2631  if ((options & PCRE_EXTENDED) != 0)
2631      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2632      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2633        {        {
2634        while (*(++ptr) != 0)        ptr++;
2635          while (*ptr != 0)
2636            {
2637          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2638            ptr++;
2639    #ifdef SUPPORT_UTF8
2640            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2641    #endif
2642            }
2643        }        }
2644      else break;      else break;
2645      }      }
# Line 2172  if ((options & PCRE_EXTENDED) != 0) Line 2675  if ((options & PCRE_EXTENDED) != 0)
2675      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2676      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2677        {        {
2678        while (*(++ptr) != 0)        ptr++;
2679          while (*ptr != 0)
2680            {
2681          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2682            ptr++;
2683    #ifdef SUPPORT_UTF8
2684            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2685    #endif
2686            }
2687        }        }
2688      else break;      else break;
2689      }      }
# Line 2185  if (*ptr == CHAR_ASTERISK || *ptr == CHA Line 2695  if (*ptr == CHAR_ASTERISK || *ptr == CHA
2695    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2696      return FALSE;      return FALSE;
2697    
2698  /* Now compare the next item with the previous opcode. If the previous is a  /* Now compare the next item with the previous opcode. First, handle cases when
2699  positive single character match, "item" either contains the character or, if  the next item is a character. */
 "item" is greater than 127 in utf8 mode, the character's bytes are in  
 utf8_char. */  
   
   
 /* Handle cases when the next item is a character. */  
2700    
2701  if (next >= 0) switch(op_code)  if (next >= 0) switch(op_code)
2702    {    {
2703    case OP_CHAR:    case OP_CHAR:
2704  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2705    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2706  #else  #else
2707    (void)(utf8_char);  /* Keep compiler happy by referencing function argument */    c = *previous;
2708  #endif  #endif
2709    return item != next;    return c != next;
2710    
2711    /* 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
2712    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
2713    high-valued characters. */    high-valued characters. */
2714    
2715    case OP_CHARNC:    case OP_CHARI:
2716  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2717    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2718    #else
2719      c = *previous;
2720  #endif  #endif
2721    if (item == next) return FALSE;    if (c == next) return FALSE;
2722  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2723    if (utf8)    if (utf8)
2724      {      {
# Line 2222  if (next >= 0) switch(op_code) Line 2729  if (next >= 0) switch(op_code)
2729  #else  #else
2730      othercase = NOTACHAR;      othercase = NOTACHAR;
2731  #endif  #endif
2732      return (unsigned int)item != othercase;      return (unsigned int)c != othercase;
2733      }      }
2734    else    else
2735  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2736    return (item != cd->fcc[next]);  /* Non-UTF-8 mode */    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2737    
2738    /* For OP_NOT, "item" must be a single-byte character. */    /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2739      opcodes are not used for multi-byte characters, because they are coded using
2740      an XCLASS instead. */
2741    
2742    case OP_NOT:    case OP_NOT:
2743    if (item == next) return TRUE;    return (c = *previous) == next;
2744    if ((options & PCRE_CASELESS) == 0) return FALSE;  
2745      case OP_NOTI:
2746      if ((c = *previous) == next) return TRUE;
2747  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2748    if (utf8)    if (utf8)
2749      {      {
# Line 2243  if (next >= 0) switch(op_code) Line 2754  if (next >= 0) switch(op_code)
2754  #else  #else
2755      othercase = NOTACHAR;      othercase = NOTACHAR;
2756  #endif  #endif
2757      return (unsigned int)item == othercase;      return (unsigned int)c == othercase;
2758      }      }
2759    else    else
2760  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2761    return (item == cd->fcc[next]);  /* Non-UTF-8 mode */    return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
2762    
2763      /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2764      When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2765    
2766    case OP_DIGIT:    case OP_DIGIT:
2767    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
# Line 2290  if (next >= 0) switch(op_code) Line 2804  if (next >= 0) switch(op_code)
2804      case 0x202f:      case 0x202f:
2805      case 0x205f:      case 0x205f:
2806      case 0x3000:      case 0x3000:
2807      return op_code != OP_HSPACE;      return op_code == OP_NOT_HSPACE;
2808      default:      default:
2809      return op_code == OP_HSPACE;      return op_code != OP_NOT_HSPACE;
2810      }      }
2811    
2812      case OP_ANYNL:
2813    case OP_VSPACE:    case OP_VSPACE:
2814    case OP_NOT_VSPACE:    case OP_NOT_VSPACE:
2815    switch(next)    switch(next)
# Line 2306  if (next >= 0) switch(op_code) Line 2821  if (next >= 0) switch(op_code)
2821      case 0x85:      case 0x85:
2822      case 0x2028:      case 0x2028:
2823      case 0x2029:      case 0x2029:
2824      return op_code != OP_VSPACE;      return op_code == OP_NOT_VSPACE;
2825      default:      default:
2826      return op_code == OP_VSPACE;      return op_code != OP_NOT_VSPACE;
2827      }      }
2828    
2829    #ifdef SUPPORT_UCP
2830      case OP_PROP:
2831      return check_char_prop(next, previous[0], previous[1], FALSE);
2832    
2833      case OP_NOTPROP:
2834      return check_char_prop(next, previous[0], previous[1], TRUE);
2835    #endif
2836    
2837    default:    default:
2838    return FALSE;    return FALSE;
2839    }    }
2840    
2841    
2842  /* Handle the case when the next item is \d, \s, etc. */  /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2843    is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2844    generated only when PCRE_UCP is *not* set, that is, when only ASCII
2845    characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2846    replaced by OP_PROP codes when PCRE_UCP is set. */
2847    
2848  switch(op_code)  switch(op_code)
2849    {    {
2850    case OP_CHAR:    case OP_CHAR:
2851    case OP_CHARNC:    case OP_CHARI:
2852  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2853    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2854    #else
2855      c = *previous;
2856  #endif  #endif
2857    switch(-next)    switch(-next)
2858      {      {
2859      case ESC_d:      case ESC_d:
2860      return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;      return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2861    
2862      case ESC_D:      case ESC_D:
2863      return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2864    
2865      case ESC_s:      case ESC_s:
2866      return item > 127 || (cd->ctypes[item] & ctype_space) == 0;      return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2867    
2868      case ESC_S:      case ESC_S:
2869      return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2870    
2871      case ESC_w:      case ESC_w:
2872      return item > 127 || (cd->ctypes[item] & ctype_word) == 0;      return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2873    
2874      case ESC_W:      case ESC_W:
2875      return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2876    
2877      case ESC_h:      case ESC_h:
2878      case ESC_H:      case ESC_H:
2879      switch(item)      switch(c)
2880        {        {
2881        case 0x09:        case 0x09:
2882        case 0x20:        case 0x20:
# Line 2375  switch(op_code) Line 2904  switch(op_code)
2904    
2905      case ESC_v:      case ESC_v:
2906      case ESC_V:      case ESC_V:
2907      switch(item)      switch(c)
2908        {        {
2909        case 0x0a:        case 0x0a:
2910        case 0x0b:        case 0x0b:
# Line 2389  switch(op_code) Line 2918  switch(op_code)
2918        return -next == ESC_v;        return -next == ESC_v;
2919        }        }
2920    
2921        /* When PCRE_UCP is set, these values get generated for \d etc. Find
2922        their substitutions and process them. The result will always be either
2923        -ESC_p or -ESC_P. Then fall through to process those values. */
2924    
2925    #ifdef SUPPORT_UCP
2926        case ESC_du:
2927        case ESC_DU:
2928        case ESC_wu:
2929        case ESC_WU:
2930        case ESC_su:
2931        case ESC_SU:
2932          {
2933          int temperrorcode = 0;
2934          ptr = substitutes[-next - ESC_DU];
2935          next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2936          if (temperrorcode != 0) return FALSE;
2937          ptr++;    /* For compatibility */
2938          }
2939        /* Fall through */
2940    
2941        case ESC_p:
2942        case ESC_P:
2943          {
2944          int ptype, pdata, errorcodeptr;
2945          BOOL negated;
2946    
2947          ptr--;      /* Make ptr point at the p or P */
2948          ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2949          if (ptype < 0) return FALSE;
2950          ptr++;      /* Point past the final curly ket */
2951    
2952          /* If the property item is optional, we have to give up. (When generated
2953          from \d etc by PCRE_UCP, this test will have been applied much earlier,
2954          to the original \d etc. At this point, ptr will point to a zero byte. */
2955    
2956          if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2957            strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2958              return FALSE;
2959    
2960          /* Do the property check. */
2961    
2962          return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2963          }
2964    #endif
2965    
2966      default:      default:
2967      return FALSE;      return FALSE;
2968      }      }
2969    
2970      /* In principle, support for Unicode properties should be integrated here as
2971      well. It means re-organizing the above code so as to get hold of the property
2972      values before switching on the op-code. However, I wonder how many patterns
2973      combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2974      these op-codes are never generated.) */
2975    
2976    case OP_DIGIT:    case OP_DIGIT:
2977    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2978           next == -ESC_h || next == -ESC_v;           next == -ESC_h || next == -ESC_v || next == -ESC_R;
2979    
2980    case OP_NOT_DIGIT:    case OP_NOT_DIGIT:
2981    return next == -ESC_d;    return next == -ESC_d;
2982    
2983    case OP_WHITESPACE:    case OP_WHITESPACE:
2984    return next == -ESC_S || next == -ESC_d || next == -ESC_w;    return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2985    
2986    case OP_NOT_WHITESPACE:    case OP_NOT_WHITESPACE:
2987    return next == -ESC_s || next == -ESC_h || next == -ESC_v;    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2988    
2989    case OP_HSPACE:    case OP_HSPACE:
2990    return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2991             next == -ESC_w || next == -ESC_v || next == -ESC_R;
2992    
2993    case OP_NOT_HSPACE:    case OP_NOT_HSPACE:
2994    return next == -ESC_h;    return next == -ESC_h;
2995    
2996    /* Can't have \S in here because VT matches \S (Perl anomaly) */    /* Can't have \S in here because VT matches \S (Perl anomaly) */
2997      case OP_ANYNL:
2998    case OP_VSPACE:    case OP_VSPACE:
2999    return next == -ESC_V || next == -ESC_d || next == -ESC_w;    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3000    
3001    case OP_NOT_VSPACE:    case OP_NOT_VSPACE:
3002    return next == -ESC_v;    return next == -ESC_v || next == -ESC_R;
3003    
3004    case OP_WORDCHAR:    case OP_WORDCHAR:
3005    return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;    return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3006             next == -ESC_v || next == -ESC_R;
3007    
3008    case OP_NOT_WORDCHAR:    case OP_NOT_WORDCHAR:
3009    return next == -ESC_w || next == -ESC_d;    return next == -ESC_w || next == -ESC_d;
# Line 2472  int greedy_default, greedy_non_default; Line 3055  int greedy_default, greedy_non_default;
3055  int firstbyte, reqbyte;  int firstbyte, reqbyte;
3056  int zeroreqbyte, zerofirstbyte;  int zeroreqbyte, zerofirstbyte;
3057  int req_caseopt, reqvary, tempreqvary;  int req_caseopt, reqvary, tempreqvary;
3058  int options = *optionsptr;  int options = *optionsptr;               /* May change dynamically */
3059  int after_manual_callout = 0;  int after_manual_callout = 0;
3060  int length_prevgroup = 0;  int length_prevgroup = 0;
3061  register int c;  register int c;
# Line 2484  BOOL inescq = FALSE; Line 3067  BOOL inescq = FALSE;
3067  BOOL groupsetfirstbyte = FALSE;  BOOL groupsetfirstbyte = FALSE;
3068  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
3069  const uschar *tempptr;  const uschar *tempptr;
3070    const uschar *nestptr = NULL;
3071  uschar *previous = NULL;  uschar *previous = NULL;
3072  uschar *previous_callout = NULL;  uschar *previous_callout = NULL;
3073  uschar *save_hwm = NULL;  uschar *save_hwm = NULL;
3074  uschar classbits[32];  uschar classbits[32];
3075    
3076    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3077    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3078    dynamically as we process the pattern. */
3079    
3080  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3081  BOOL class_utf8;  BOOL class_utf8;
3082  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8 = (options & PCRE_UTF8) != 0;
# Line 2500  BOOL utf8 = FALSE; Line 3088  BOOL utf8 = FALSE;
3088  uschar *utf8_char = NULL;  uschar *utf8_char = NULL;
3089  #endif  #endif
3090    
3091  #ifdef DEBUG  #ifdef PCRE_DEBUG
3092  if (lengthptr != NULL) DPRINTF((">> start branch\n"));  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3093  #endif  #endif
3094    
# Line 2554  for (;; ptr++) Line 3142  for (;; ptr++)
3142    
3143    c = *ptr;    c = *ptr;
3144    
3145      /* If we are at the end of a nested substitution, revert to the outer level
3146      string. Nesting only happens one level deep. */
3147    
3148      if (c == 0 && nestptr != NULL)
3149        {
3150        ptr = nestptr;
3151        nestptr = NULL;
3152        c = *ptr;
3153        }
3154    
3155    /* If we are in the pre-compile phase, accumulate the length used for the    /* If we are in the pre-compile phase, accumulate the length used for the
3156    previous cycle of this loop. */    previous cycle of this loop. */
3157    
3158    if (lengthptr != NULL)    if (lengthptr != NULL)
3159      {      {
3160  #ifdef DEBUG  #ifdef PCRE_DEBUG
3161      if (code > cd->hwm) cd->hwm = code;                 /* High water info */      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3162  #endif  #endif
3163      if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */      if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3164        {        {
3165        *errorcodeptr = ERR52;        *errorcodeptr = ERR52;
3166        goto FAILED;        goto FAILED;
# Line 2584  for (;; ptr++) Line 3182  for (;; ptr++)
3182        goto FAILED;        goto FAILED;
3183        }        }
3184    
3185      *lengthptr += code - last_code;      *lengthptr += (int)(code - last_code);
3186      DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));      DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3187    
3188      /* 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
# Line 2611  for (;; ptr++) Line 3209  for (;; ptr++)
3209    /* 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
3210    reference list. */    reference list. */
3211    
3212    else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3213      {      {
3214      *errorcodeptr = ERR52;      *errorcodeptr = ERR52;
3215      goto FAILED;      goto FAILED;
# Line 2659  for (;; ptr++) Line 3257  for (;; ptr++)
3257      previous_callout = NULL;      previous_callout = NULL;
3258      }      }
3259    
3260    /* In extended mode, skip white space and comments */    /* In extended mode, skip white space and comments. */
3261    
3262    if ((options & PCRE_EXTENDED) != 0)    if ((options & PCRE_EXTENDED) != 0)
3263      {      {
3264      if ((cd->ctypes[c] & ctype_space) != 0) continue;      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3265      if (c == CHAR_NUMBER_SIGN)      if (c == CHAR_NUMBER_SIGN)
3266        {        {
3267        while (*(++ptr) != 0)        ptr++;
3268          while (*ptr != 0)
3269          {          {
3270          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3271            ptr++;
3272    #ifdef SUPPORT_UTF8
3273            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3274    #endif
3275          }          }
3276        if (*ptr != 0) continue;        if (*ptr != 0) continue;
3277    
# Line 2702  for (;; ptr++) Line 3305  for (;; ptr++)
3305          *errorcodeptr = ERR20;          *errorcodeptr = ERR20;
3306          goto FAILED;          goto FAILED;
3307          }          }
3308        *lengthptr += code - last_code;   /* To include callout length */        *lengthptr += (int)(code - last_code);   /* To include callout length */
3309        DPRINTF((">> end branch\n"));        DPRINTF((">> end branch\n"));
3310        }        }
3311      return TRUE;      return TRUE;
# Line 2713  for (;; ptr++) Line 3316  for (;; ptr++)
3316      the setting of any following char as a first character. */      the setting of any following char as a first character. */
3317    
3318      case CHAR_CIRCUMFLEX_ACCENT:      case CHAR_CIRCUMFLEX_ACCENT:
3319        previous = NULL;
3320      if ((options & PCRE_MULTILINE) != 0)      if ((options & PCRE_MULTILINE) != 0)
3321        {        {
3322        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3323          *code++ = OP_CIRCM;
3324        }        }
3325      previous = NULL;      else *code++ = OP_CIRC;
     *code++ = OP_CIRC;  
3326      break;      break;
3327    
3328      case CHAR_DOLLAR_SIGN:      case CHAR_DOLLAR_SIGN:
3329      previous = NULL;      previous = NULL;
3330      *code++ = OP_DOLL;      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3331      break;      break;
3332    
3333      /* 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 2907  for (;; ptr++) Line 3511  for (;; ptr++)
3511            ptr++;            ptr++;
3512            }            }
3513    
3514          posix_class = check_posix_name(ptr, tempptr - ptr);          posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3515          if (posix_class < 0)          if (posix_class < 0)
3516            {            {
3517            *errorcodeptr = ERR30;            *errorcodeptr = ERR30;
# Line 2921  for (;; ptr++) Line 3525  for (;; ptr++)
3525          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3526            posix_class = 0;            posix_class = 0;
3527    
3528          /* We build the bit map for the POSIX class in a chunk of local store          /* When PCRE_UCP is set, some of the POSIX classes are converted to
3529          because we may be adding and subtracting from it, and we don't want to          different escape sequences that use Unicode properties. */
3530          subtract bits that may be in the main map already. At the end we or the  
3531          result into the bit map that is being built. */  #ifdef SUPPORT_UCP
3532            if ((options & PCRE_UCP) != 0)
3533              {
3534              int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3535              if (posix_substitutes[pc] != NULL)
3536                {
3537                nestptr = tempptr + 1;
3538                ptr = posix_substitutes[pc] - 1;
3539                continue;
3540                }
3541              }
3542    #endif
3543            /* In the non-UCP case, we build the bit map for the POSIX class in a
3544            chunk of local store because we may be adding and subtracting from it,
3545            and we don't want to subtract bits that may be in the main map already.
3546            At the end we or the result into the bit map that is being built. */
3547    
3548          posix_class *= 3;          posix_class *= 3;
3549    
# Line 2968  for (;; ptr++) Line 3587  for (;; ptr++)
3587    
3588        /* Backslash may introduce a single character, or it may introduce one        /* Backslash may introduce a single character, or it may introduce one
3589        of the specials, which just set a flag. The sequence \b is a special        of the specials, which just set a flag. The sequence \b is a special
3590        case. Inside a class (and only there) it is treated as backspace.        case. Inside a class (and only there) it is treated as backspace. We
3591        Elsewhere it marks a word boundary. Other escapes have preset maps ready        assume that other escapes have more than one character in them, so set
3592        to 'or' into the one we are building. We assume they have more than one        class_charcount bigger than one. Unrecognized escapes fall through and
3593        character in them, so set class_charcount bigger than one. */        are either treated as literal characters (by default), or are faulted if
3594          PCRE_EXTRA is set. */
3595    
3596        if (c == CHAR_BACKSLASH)        if (c == CHAR_BACKSLASH)
3597          {          {
3598          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3599          if (*errorcodeptr != 0) goto FAILED;          if (*errorcodeptr != 0) goto FAILED;
3600    
3601          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 */
         else if (-c == ESC_X) c = CHAR_X;   /* \X is literal X in a class */  
         else if (-c == ESC_R) c = CHAR_R;   /* \R is literal R in a class */  
3602          else if (-c == ESC_Q)            /* Handle start of quoted string */          else if (-c == ESC_Q)            /* Handle start of quoted string */
3603            {            {
3604            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
# Line 2997  for (;; ptr++) Line 3615  for (;; ptr++)
3615            register const uschar *cbits = cd->cbits;            register const uschar *cbits = cd->cbits;
3616            class_charcount += 2;     /* Greater than 1 is what matters */            class_charcount += 2;     /* Greater than 1 is what matters */
3617    
3618            /* Save time by not doing this in the pre-compile phase. */            switch (-c)
   
           if (lengthptr == NULL) switch (-c)  
3619              {              {
3620    #ifdef SUPPORT_UCP
3621                case ESC_du:     /* These are the values given for \d etc */
3622                case ESC_DU:     /* when PCRE_UCP is set. We replace the */
3623                case ESC_wu:     /* escape sequence with an appropriate \p */
3624                case ESC_WU:     /* or \P to test Unicode properties instead */
3625                case ESC_su:     /* of the default ASCII testing. */
3626                case ESC_SU:
3627                nestptr = ptr;
3628                ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
3629                class_charcount -= 2;                /* Undo! */
3630                continue;
3631    #endif
3632              case ESC_d:              case ESC_d:
3633              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3634              continue;              continue;
# Line 3019  for (;; ptr++) Line 3647  for (;; ptr++)
3647              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3648              continue;              continue;
3649    
3650                /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3651                if it was previously set by something earlier in the character
3652                class. */
3653    
3654              case ESC_s:              case ESC_s:
3655              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];              classbits[0] |= cbits[cbit_space];
3656              classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= cbits[cbit_space+1] & ~0x08;
3657                for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3658              continue;              continue;
3659    
3660              case ESC_S:              case ESC_S:
# Line 3030  for (;; ptr++) Line 3663  for (;; ptr++)
3663              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
3664              continue;              continue;
3665    
3666              default:    /* Not recognized; fall through */              case ESC_h:
             break;      /* Need "default" setting to stop compiler warning. */  
             }  
   
           /* In the pre-compile phase, just do the recognition. */  
   
           else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||  
                    c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;  
   
           /* We need to deal with \H, \h, \V, and \v in both phases because  
           they use extra memory. */  
   
           if (-c == ESC_h)  
             {  
3667              SETBIT(classbits, 0x09); /* VT */              SETBIT(classbits, 0x09); /* VT */
3668              SETBIT(classbits, 0x20); /* SPACE */              SETBIT(classbits, 0x20); /* SPACE */
3669              SETBIT(classbits, 0xa0); /* NSBP */              SETBIT(classbits, 0xa0); /* NSBP */
# Line 3067  for (;; ptr++) Line 3687  for (;; ptr++)
3687                }                }
3688  #endif  #endif
3689              continue;              continue;
             }  
3690    
3691            if (-c == ESC_H)              case ESC_H:
             {  
3692              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3693                {                {
3694                int x = 0xff;                int x = 0xff;
# Line 3112  for (;; ptr++) Line 3730  for (;; ptr++)
3730                }                }
3731  #endif  #endif
3732              continue;              continue;
             }  
3733    
3734            if (-c == ESC_v)              case ESC_v:
             {  
3735              SETBIT(classbits, 0x0a); /* LF */              SETBIT(classbits, 0x0a); /* LF */
3736              SETBIT(classbits, 0x0b); /* VT */              SETBIT(classbits, 0x0b); /* VT */
3737              SETBIT(classbits, 0x0c); /* FF */              SETBIT(classbits, 0x0c); /* FF */
# Line 3131  for (;; ptr++) Line 3747  for (;; ptr++)
3747                }                }
3748  #endif  #endif
3749              continue;              continue;
             }  
3750    
3751            if (-c == ESC_V)              case ESC_V:
             {  
3752              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3753                {                {
3754                int x = 0xff;                int x = 0xff;
# Line 3164  for (;; ptr++) Line 3778  for (;; ptr++)
3778                }                }
3779  #endif  #endif
3780              continue;              continue;
             }  
   
           /* We need to deal with \P and \p in both phases. */  
3781    
3782  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3783            if (-c == ESC_p || -c == ESC_P)              case ESC_p:
3784              {              case ESC_P:
3785              BOOL negated;                {
3786              int pdata;                BOOL negated;
3787              int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);                int pdata;
3788              if (ptype < 0) goto FAILED;                int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3789              class_utf8 = TRUE;                if (ptype < 0) goto FAILED;
3790              *class_utf8data++ = ((-c == ESC_p) != negated)?                class_utf8 = TRUE;
3791                XCL_PROP : XCL_NOTPROP;                *class_utf8data++ = ((-c == ESC_p) != negated)?
3792              *class_utf8data++ = ptype;                  XCL_PROP : XCL_NOTPROP;
3793              *class_utf8data++ = pdata;                *class_utf8data++ = ptype;
3794              class_charcount -= 2;   /* Not a < 256 character */                *class_utf8data++ = pdata;
3795              continue;                class_charcount -= 2;   /* Not a < 256 character */
3796              }                continue;
3797                  }
3798  #endif  #endif
3799            /* Unrecognized escapes are faulted if PCRE is running in its              /* Unrecognized escapes are faulted if PCRE is running in its
3800            strict mode. By default, for compatibility with Perl, they are              strict mode. By default, for compatibility with Perl, they are
3801            treated as literals. */              treated as literals. */
3802    
3803            if ((options & PCRE_EXTRA) != 0)              default:
3804              {              if ((options & PCRE_EXTRA) != 0)
3805              *errorcodeptr = ERR7;                {
3806              goto FAILED;                *errorcodeptr = ERR7;
3807                  goto FAILED;
3808                  }
3809                class_charcount -= 2;  /* Undo the default count from above */
3810                c = *ptr;              /* Get the final character and fall through */
3811                break;
3812              }              }
   
           class_charcount -= 2;  /* Undo the default count from above */  
           c = *ptr;              /* Get the final character and fall through */  
3813            }            }
3814    
3815          /* Fall through if we have a single character (c >= 0). This may be          /* Fall through if we have a single character (c >= 0). This may be
# Line 3265  for (;; ptr++) Line 3879  for (;; ptr++)
3879            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3880            if (*errorcodeptr != 0) goto FAILED;            if (*errorcodeptr != 0) goto FAILED;
3881    
3882            /* \b is backspace; \X is literal X; \R is literal R; any other            /* \b is backspace; any other special means the '-' was literal */
           special means the '-' was literal */  
3883    
3884            if (d < 0)            if (d < 0)
3885              {              {
3886              if (d == -ESC_b) d = CHAR_BS;              if (d == -ESC_b) d = CHAR_BS; else
             else if (d == -ESC_X) d = CHAR_X;  
             else if (d == -ESC_R) d = CHAR_R; else  
3887                {                {
3888                ptr = oldptr;                ptr = oldptr;
3889                goto LONE_SINGLE_CHARACTER;  /* A few lines below */                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
# Line 3438  for (;; ptr++) Line 4049  for (;; ptr++)
4049          }          }
4050        }        }
4051    
4052      /* Loop until ']' reached. This "while" is the end of the "do" above. */      /* Loop until ']' reached. This "while" is the end of the "do" far above.
4053        If we are at the end of an internal nested string, revert to the outer
4054        string. */
4055    
4056        while (((c = *(++ptr)) != 0 ||
4057               (nestptr != NULL &&
4058                 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4059               (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4060    
4061      while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));      /* Check for missing terminating ']' */
4062    
4063      if (c == 0)                          /* Missing terminating ']' */      if (c == 0)
4064        {        {
4065        *errorcodeptr = ERR6;        *errorcodeptr = ERR6;
4066        goto FAILED;        goto FAILED;
4067        }        }
4068    
   
 /* This code has been disabled because it would mean that \s counts as  
 an explicit \r or \n reference, and that's not really what is wanted. Now  
 we set the flag only if there is a literal "\r" or "\n" in the class. */  
   
 #if 0  
     /* Remember whether \r or \n are in this class */  
   
     if (negate_class)  
       {  
       if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;  
       }  
     else  
       {  
       if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;  
       }  
 #endif  
   
   
4069      /* If class_charcount is 1, we saw precisely one character whose value is      /* If class_charcount is 1, we saw precisely one character whose value is
4070      less than 256. As long as there were no characters >= 128 and there was no      less than 256. As long as there were no characters >= 128 and there was no
4071      use of \p or \P, in other words, no use of any XCLASS features, we can      use of \p or \P, in other words, no use of any XCLASS features, we can
# Line 3474  we set the flag only if there is a liter Line 4073  we set the flag only if there is a liter
4073    
4074      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
4075      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4076      operate on single-bytes only. This is an historical hangover. Maybe one day      operate on single-bytes characters only. This is an historical hangover.
4077      we can tidy these opcodes to handle multi-byte characters.      Maybe one day we can tidy these opcodes to handle multi-byte characters.
4078    
4079      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
4080      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.
4081      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
4082      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
4083      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
4084      reqbyte, save the previous value for reinstating. */      of reqbyte, save the previous value for reinstating. */
4085    
4086  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4087      if (class_charcount == 1 && !class_utf8 &&      if (class_charcount == 1 && !class_utf8 &&
# Line 3493  we set the flag only if there is a liter Line 4092  we set the flag only if there is a liter
4092        {        {
4093        zeroreqbyte = reqbyte;        zeroreqbyte = reqbyte;
4094    
4095        /* The OP_NOT opcode works on one-byte characters only. */        /* The OP_NOT[I] opcodes work on one-byte characters only. */
4096    
4097        if (negate_class)        if (negate_class)
4098          {          {
4099          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4100          zerofirstbyte = firstbyte;          zerofirstbyte = firstbyte;
4101          *code++ = OP_NOT;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4102          *code++ = class_lastchar;          *code++ = class_lastchar;
4103          break;          break;
4104          }          }
# Line 3530  we set the flag only if there is a liter Line 4129  we set the flag only if there is a liter
4129    
4130      /* If there are characters with values > 255, we have to compile an      /* If there are characters with values > 255, we have to compile an
4131      extended class, with its own opcode, unless there was a negated special      extended class, with its own opcode, unless there was a negated special
4132      such as \S in the class, because in that case all characters > 255 are in      such as \S in the class, and PCRE_UCP is not set, because in that case all
4133      the class, so any that were explicitly given as well can be ignored. If      characters > 255 are in the class, so any that were explicitly given as
4134      (when there are explicit characters > 255 that must be listed) there are no      well can be ignored. If (when there are explicit characters > 255 that must
4135      characters < 256, we can omit the bitmap in the actual compiled code. */      be listed) there are no characters < 256, we can omit the bitmap in the
4136        actual compiled code. */
4137    
4138  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4139      if (class_utf8 && !should_flip_negation)      if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4140        {        {
4141        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
4142        *code++ = OP_XCLASS;        *code++ = OP_XCLASS;
# Line 3562  we set the flag only if there is a liter Line 4162  we set the flag only if there is a liter
4162        }        }
4163  #endif  #endif
4164    
4165      /* If there are no characters > 255, set the opcode to OP_CLASS or      /* If there are no characters > 255, or they are all to be included or
4166      OP_NCLASS, depending on whether the whole class was negated and whether      excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
4167      there were negative specials such as \S in the class. Then copy the 32-byte      whole class was negated and whether there were negative specials such as \S
4168      map into the code vector, negating it if necessary. */      (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4169        negating it if necessary. */
4170    
4171      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4172      if (negate_class)      if (negate_class)
# Line 3625  we set the flag only if there is a liter Line 4226  we set the flag only if there is a liter
4226      op_type = 0;                    /* Default single-char op codes */      op_type = 0;                    /* Default single-char op codes */
4227      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
4228    
4229      /* 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
4230      for an inserted OP_ONCE for the additional '+' extension. */      insert something before it. */
4231    
4232      tempcode = previous;      tempcode = previous;
4233    
# Line 3648  we set the flag only if there is a liter Line 4249  we set the flag only if there is a liter
4249        ptr++;        ptr++;
4250        }        }
4251      else repeat_type = greedy_default;      else repeat_type = greedy_default;
4252    
4253        /* If previous was a recursion call, wrap it in atomic brackets so that
4254        previous becomes the atomic group. All recursions were so wrapped in the
4255        past, but it no longer happens for non-repeated recursions. In fact, the
4256        repeated ones could be re-implemented independently so as not to need this,
4257        but for the moment we rely on the code for repeating groups. */
4258    
4259        if (*previous == OP_RECURSE)
4260          {
4261          memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4262          *previous = OP_ONCE;
4263          PUT(previous, 1, 2 + 2*LINK_SIZE);
4264          previous[2 + 2*LINK_SIZE] = OP_KET;
4265          PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4266          code += 2 + 2 * LINK_SIZE;
4267          length_prevgroup = 3 + 3*LINK_SIZE;
4268    
4269          /* When actually compiling, we need to check whether this was a forward
4270          reference, and if so, adjust the offset. */
4271    
4272          if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4273            {
4274            int offset = GET(cd->hwm, -LINK_SIZE);
4275            if (offset == previous + 1 - cd->start_code)
4276              PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4277            }
4278          }
4279    
4280        /* Now handle repetition for the different types of item. */
4281    
4282      /* If previous was a character match, abolish the item and generate a      /* If previous was a character match, abolish the item and generate a
4283      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
# Line 3655  we set the flag only if there is a liter Line 4285  we set the flag only if there is a liter
4285      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
4286      instead.  */      instead.  */
4287    
4288      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (*previous == OP_CHAR || *previous == OP_CHARI)
4289        {        {
4290          op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4291    
4292        /* 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
4293        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
4294        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 3689  we set the flag only if there is a liter Line 4321  we set the flag only if there is a liter
4321    
4322        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4323            repeat_max < 0 &&            repeat_max < 0 &&
4324            check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
             options, cd))  
4325          {          {
4326          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4327          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3702  we set the flag only if there is a liter Line 4333  we set the flag only if there is a liter
4333      /* If previous was a single negated character ([^a] or similar), we use      /* If previous was a single negated character ([^a] or similar), we use
4334      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-
4335      character repeats by setting opt_type to add a suitable offset into      character repeats by setting opt_type to add a suitable offset into
4336      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
4337      currently used only for single-byte chars. */      are currently used only for single-byte chars. */
4338    
4339      else if (*previous == OP_NOT)      else if (*previous == OP_NOT || *previous == OP_NOTI)
4340        {        {
4341        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4342        c = previous[1];        c = previous[1];
4343        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4344            repeat_max < 0 &&            repeat_max < 0 &&
4345            check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4346          {          {
4347          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4348          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3735  we set the flag only if there is a liter Line 4366  we set the flag only if there is a liter
4366    
4367        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4368            repeat_max < 0 &&            repeat_max < 0 &&
4369            check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4370          {          {
4371          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4372          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3757  we set the flag only if there is a liter Line 4388  we set the flag only if there is a liter
4388    
4389        if (repeat_max == 0) goto END_REPEAT;        if (repeat_max == 0) goto END_REPEAT;
4390    
4391          /*--------------------------------------------------------------------*/
4392          /* This code is obsolete from release 8.00; the restriction was finally
4393          removed: */
4394    
4395        /* All real repeats make it impossible to handle partial matching (maybe        /* All real repeats make it impossible to handle partial matching (maybe
4396        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4397    
4398        if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4399          /*--------------------------------------------------------------------*/
4400    
4401        /* Combine the op_type with the repeat_type */        /* Combine the op_type with the repeat_type */
4402    
# Line 3899  we set the flag only if there is a liter Line 4535  we set the flag only if there is a liter
4535  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4536               *previous == OP_XCLASS ||               *previous == OP_XCLASS ||
4537  #endif  #endif
4538               *previous == OP_REF)               *previous == OP_REF ||
4539                 *previous == OP_REFI)
4540        {        {
4541        if (repeat_max == 0)        if (repeat_max == 0)
4542          {          {
# Line 3907  we set the flag only if there is a liter Line 4544  we set the flag only if there is a liter
4544          goto END_REPEAT;          goto END_REPEAT;
4545          }          }
4546    
4547          /*--------------------------------------------------------------------*/
4548          /* This code is obsolete from release 8.00; the restriction was finally
4549          removed: */
4550    
4551        /* All real repeats make it impossible to handle partial matching (maybe        /* All real repeats make it impossible to handle partial matching (maybe
4552        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4553    
4554        if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4555          /*--------------------------------------------------------------------*/
4556    
4557        if (repeat_min == 0 && repeat_max == -1)        if (repeat_min == 0 && repeat_max == -1)
4558          *code++ = OP_CRSTAR + repeat_type;          *code++ = OP_CRSTAR + repeat_type;
# Line 3928  we set the flag only if there is a liter Line 4570  we set the flag only if there is a liter
4570        }        }
4571    
4572      /* 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
4573      cases. */      cases. Note that at this point we can encounter only the "basic" bracket
4574        opcodes such as BRA and CBRA, as this is the place where they get converted
4575        into the more special varieties such as BRAPOS and SBRA. A test for >=
4576        OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4577        ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4578        repetition of assertions, but now it does, for Perl compatibility. */
4579    
4580      else if (*previous == OP_BRA  || *previous == OP_CBRA ||      else if (*previous >= OP_ASSERT && *previous <= OP_COND)
              *previous == OP_ONCE || *previous == OP_COND)  
4581        {        {
4582        register int i;        register int i;
4583        int ketoffset = 0;        int len = (int)(code - previous);
       int len = code - previous;  
4584        uschar *bralink = NULL;        uschar *bralink = NULL;
4585          uschar *brazeroptr = NULL;
4586        /* Repeating a DEFINE group is pointless */  
4587          /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4588          we just ignore the repeat. */
4589    
4590        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4591          {          goto END_REPEAT;
         *errorcodeptr = ERR55;  
         goto FAILED;  
         }  
4592    
4593        /* If the maximum repeat count is unlimited, find the end of the bracket        /* There is no sense in actually repeating assertions. The only potential
4594        by scanning through from the start, and compute the offset back to it        use of repetition is in cases when the assertion is optional. Therefore,
4595        from the current code pointer. There may be an OP_OPT setting following        if the minimum is greater than zero, just ignore the repeat. If the
4596        the final KET, so we can't find the end just by going back from the code        maximum is not not zero or one, set it to 1. */
4597        pointer. */  
4598          if (*previous < OP_ONCE)    /* Assertion */
4599        if (repeat_max == -1)          {
4600          {          if (repeat_min > 0) goto END_REPEAT;
4601          register uschar *ket = previous;          if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4602          do ket += GET(ket, 1); while (*ket != OP_KET);          }
         ketoffset = code - ket;  
         }  
4603    
4604        /* 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
4605        OP_BRAZERO in front of it, and because the group appears once in the        OP_BRAZERO in front of it, and because the group appears once in the
# Line 3977  we set the flag only if there is a liter Line 4619  we set the flag only if there is a liter
4619          **   goto END_REPEAT;          **   goto END_REPEAT;
4620          **   }          **   }
4621    
4622          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
4623          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
4624          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
4625          groups are referenced, we cannot do this selectively.          don't have a list of which groups are referenced, we cannot do this
4626            selectively.
4627    
4628          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
4629          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 4000  we set the flag only if there is a liter Line 4643  we set the flag only if there is a liter
4643              *previous++ = OP_SKIPZERO;              *previous++ = OP_SKIPZERO;
4644              goto END_REPEAT;              goto END_REPEAT;
4645              }              }
4646              brazeroptr = previous;    /* Save for possessive optimizing */
4647            *previous++ = OP_BRAZERO + repeat_type;            *previous++ = OP_BRAZERO + repeat_type;
4648            }            }
4649    
# Line 4024  we set the flag only if there is a liter Line 4668  we set the flag only if there is a liter
4668            /* We chain together the bracket offset fields that have to be            /* We chain together the bracket offset fields that have to be
4669            filled in later when the ends of the brackets are reached. */            filled in later when the ends of the brackets are reached. */
4670    
4671            offset = (bralink == NULL)? 0 : previous - bralink;            offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4672            bralink = previous;            bralink = previous;
4673            PUTINC(previous, 0, offset);            PUTINC(previous, 0, offset);
4674            }            }
# Line 4045  we set the flag only if there is a liter Line 4689  we set the flag only if there is a liter
4689            {            {
4690            /* In the pre-compile phase, we don't actually do the replication. We            /* In the pre-compile phase, we don't actually do the replication. We
4691            just adjust the length as if we had. Do some paranoid checks for            just adjust the length as if we had. Do some paranoid checks for
4692            potential integer overflow. */            potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4693              integer type when available, otherwise double. */
4694    
4695            if (lengthptr != NULL)            if (lengthptr != NULL)
4696              {              {
4697              int delta = (repeat_min - 1)*length_prevgroup;              int delta = (repeat_min - 1)*length_prevgroup;
4698              if ((double)(repeat_min - 1)*(double)length_prevgroup >              if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4699                                                              (double)INT_MAX ||                    (INT64_OR_DOUBLE)length_prevgroup >
4700                        (INT64_OR_DOUBLE)INT_MAX ||
4701                  OFLOW_MAX - *lengthptr < delta)                  OFLOW_MAX - *lengthptr < delta)
4702                {                {
4703                *errorcodeptr = ERR20;                *errorcodeptr = ERR20;
# Line 4097  we set the flag only if there is a liter Line 4743  we set the flag only if there is a liter
4743          just adjust the length as if we had. For each repetition we must add 1          just adjust the length as if we had. For each repetition we must add 1
4744          to the length for BRAZERO and for all but the last repetition we must          to the length for BRAZERO and for all but the last repetition we must
4745          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some          add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4746          paranoid checks to avoid integer overflow. */          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4747            a 64-bit integer type when available, otherwise double. */
4748    
4749          if (lengthptr != NULL && repeat_max > 0)          if (lengthptr != NULL && repeat_max > 0)
4750            {            {
4751            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4752                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */
4753            if ((double)repeat_max *            if ((INT64_OR_DOUBLE)repeat_max *
4754                  (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)                  (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4755                    > (double)INT_MAX ||                    > (INT64_OR_DOUBLE)INT_MAX ||
4756                OFLOW_MAX - *lengthptr < delta)                OFLOW_MAX - *lengthptr < delta)
4757              {              {
4758              *errorcodeptr = ERR20;              *errorcodeptr = ERR20;
# Line 4130  we set the flag only if there is a liter Line 4777  we set the flag only if there is a liter
4777              {              {
4778              int offset;              int offset;
4779              *code++ = OP_BRA;              *code++ = OP_BRA;
4780              offset = (bralink == NULL)? 0 : code - bralink;              offset = (bralink == NULL)? 0 : (int)(code - bralink);
4781              bralink = code;              bralink = code;
4782              PUTINC(code, 0, offset);              PUTINC(code, 0, offset);
4783              }              }
# Line 4151  we set the flag only if there is a liter Line 4798  we set the flag only if there is a liter
4798          while (bralink != NULL)          while (bralink != NULL)
4799            {            {
4800            int oldlinkoffset;            int oldlinkoffset;
4801            int offset = code - bralink + 1;            int offset = (int)(code - bralink + 1);
4802            uschar *bra = code - offset;            uschar *bra = code - offset;
4803            oldlinkoffset = GET(bra, 1);            oldlinkoffset = GET(bra, 1);
4804            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
# Line 4161  we set the flag only if there is a liter Line 4808  we set the flag only if there is a liter
4808            }            }
4809          }          }
4810    
4811        /* 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
4812        can't just offset backwards from the current code point, because we        ONCE brackets, that's all we need to do. However, possessively repeated
4813        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
4814        correct offset was computed above.        behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4815          deal with possessive ONCEs specially.
4816    
4817          Otherwise, if the quantifier was possessive, we convert the BRA code to
4818          the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4819          at runtime to detect this kind of subpattern at both the start and at the
4820          end.) The use of special opcodes makes it possible to reduce greatly the
4821          stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4822          convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4823          the default action below, of wrapping everything inside atomic brackets,
4824          does not happen.
4825    
4826        Then, when we are doing the actual compile phase, check to see whether        Then, when we are doing the actual compile phase, check to see whether
4827        this group is a non-atomic one that could match an empty string. If so,        this group is one that could match an empty string. If so, convert the
4828        convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so        initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so that runtime
4829        that runtime checking can be done. [This check is also applied to        checking can be done. [This check is also applied to ONCE groups at
4830        atomic groups at runtime, but in a different way.] */        runtime, but in a different way.] */
4831    
4832        else        else
4833          {          {
4834          uschar *ketcode = code - ketoffset;          uschar *ketcode = code - 1 - LINK_SIZE;
4835          uschar *bracode = ketcode - GET(ketcode, 1);          uschar *bracode = ketcode - GET(ketcode, 1);
4836          *ketcode = OP_KETRMAX + repeat_type;  
4837          if (lengthptr == NULL && *bracode != OP_ONCE)          if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4838            if (*bracode == OP_ONCE)
4839              *ketcode = OP_KETRMAX + repeat_type;
4840            else
4841            {            {
4842            uschar *scode = bracode;            if (possessive_quantifier)
4843            do              {
4844                *bracode += 1;                   /* Switch to xxxPOS opcodes */
4845                *ketcode = OP_KETRPOS;
4846                if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4847                possessive_quantifier = FALSE;
4848                }
4849              else *ketcode = OP_KETRMAX + repeat_type;
4850    
4851              if (lengthptr == NULL)
4852              {              {
4853              if (could_be_empty_branch(scode, ketcode, utf8))              uschar *scode = bracode;
4854                do
4855                {                {
4856                *bracode += OP_SBRA - OP_BRA;                if (could_be_empty_branch(scode, ketcode, utf8, cd))
4857                break;                  {
4858                    *bracode += OP_SBRA - OP_BRA;
4859                    break;
4860                    }
4861                  scode += GET(scode, 1);
4862                }                }
4863              scode += GET(scode, 1);              while (*scode == OP_ALT);
4864              }              }
           while (*scode == OP_ALT);  
4865            }            }
4866          }          }
4867        }        }
# Line 4210  we set the flag only if there is a liter Line 4882  we set the flag only if there is a liter
4882        }        }
4883    
4884      /* If the character following a repeat is '+', or if certain optimization      /* If the character following a repeat is '+', or if certain optimization
4885      tests above succeeded, possessive_quantifier is TRUE. For some of the      tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4886      simpler opcodes, there is an special alternative opcode for this. For      there are special alternative opcodes for this case. For anything else, we
4887      anything else, we wrap the entire repeated item inside OP_ONCE brackets.      wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4888      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
4889      but the special opcodes can optimize it a bit. The repeated item starts at      special opcodes can optimize it.
4890      tempcode, not at previous, which might be the first part of a string whose  
4891      (former) last char we repeated.      Possessively repeated subpatterns have already been handled in the code
4892        just above, so possessive_quantifier is always FALSE for them at this
4893        stage.
4894    
4895        Note that the repeated item starts at tempcode, not at previous, which
4896        might be the first part of a string whose (former) last char we repeated.
4897    
4898      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
4899      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 4225  we set the flag only if there is a liter Line 4902  we set the flag only if there is a liter
4902      if (possessive_quantifier)      if (possessive_quantifier)
4903        {        {
4904        int len;        int len;
4905        if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||  
4906            *tempcode == OP_NOTEXACT)        if (*tempcode == OP_TYPEEXACT)
4907          tempcode += _pcre_OP_lengths[*tempcode] +          tempcode += _pcre_OP_lengths[*tempcode] +
4908            ((*tempcode == OP_TYPEEXACT &&            ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4909               (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);  
4910        len = code - tempcode;        else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4911            {
4912            tempcode += _pcre_OP_lengths[*tempcode];
4913    #ifdef SUPPORT_UTF8
4914            if (utf8 && tempcode[-1] >= 0xc0)
4915              tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4916    #endif
4917            }
4918    
4919          len = (int)(code - tempcode);
4920        if (len > 0) switch (*tempcode)        if (len > 0) switch (*tempcode)
4921          {          {
4922          case OP_STAR:  *tempcode = OP_POSSTAR; break;          case OP_STAR:  *tempcode = OP_POSSTAR; break;
# Line 4238  we set the flag only if there is a liter Line 4924  we set the flag only if there is a liter
4924          case OP_QUERY: *tempcode = OP_POSQUERY; break;          case OP_QUERY: *tempcode = OP_POSQUERY; break;
4925          case OP_UPTO:  *tempcode = OP_POSUPTO; break;          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
4926    
4927          case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;          case OP_STARI:  *tempcode = OP_POSSTARI; break;
4928          case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;          case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
4929          case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;          case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4930          case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;          case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
4931    
4932          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
4933          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
4934          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4935          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
4936    
4937            case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
4938            case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
4939            case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4940            case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
4941    
4942            case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
4943            case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
4944            case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4945            case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
4946    
4947            /* Because we are moving code along, we must ensure that any
4948            pending recursive references are updated. */
4949    
4950          default:          default:
4951            *code = OP_END;
4952            adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4953          memmove(tempcode + 1+LINK_SIZE, tempcode, len);          memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4954          code += 1 + LINK_SIZE;          code += 1 + LINK_SIZE;
4955          len += 1 + LINK_SIZE;          len += 1 + LINK_SIZE;
# Line 4284  we set the flag only if there is a liter Line 4985  we set the flag only if there is a liter
4985    
4986      /* First deal with various "verbs" that can be introduced by '*'. */      /* First deal with various "verbs" that can be introduced by '*'. */
4987    
4988      if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)      if (*(++ptr) == CHAR_ASTERISK &&
4989             ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4990        {        {
4991        int i, namelen;        int i, namelen;
4992          int arglen = 0;
4993        const char *vn = verbnames;        const char *vn = verbnames;
4994        const uschar *name = ++ptr;        const uschar *name = ptr + 1;
4995          const uschar *arg = NULL;
4996        previous = NULL;        previous = NULL;
4997        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4998          namelen = (int)(ptr - name);
4999    
5000        if (*ptr == CHAR_COLON)        if (*ptr == CHAR_COLON)
5001          {          {
5002          *errorcodeptr = ERR59;   /* Not supported */          arg = ++ptr;
5003          goto FAILED;          while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
5004              || *ptr == '_') ptr++;
5005            arglen = (int)(ptr - arg);
5006          }          }
5007    
5008        if (*ptr != CHAR_RIGHT_PARENTHESIS)        if (*ptr != CHAR_RIGHT_PARENTHESIS)
5009          {          {
5010          *errorcodeptr = ERR60;          *errorcodeptr = ERR60;
5011          goto FAILED;          goto FAILED;
5012          }          }
5013        namelen = ptr - name;  
5014          /* Scan the table of verb names */
5015    
5016        for (i = 0; i < verbcount; i++)        for (i = 0; i < verbcount; i++)
5017          {          {
5018          if (namelen == verbs[i].len &&          if (namelen == verbs[i].len &&
5019              strncmp((char *)name, vn, namelen) == 0)              strncmp((char *)name, vn, namelen) == 0)
5020            {            {
5021            *code = verbs[i].op;            /* Check for open captures before ACCEPT and convert it to
5022            if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;            ASSERT_ACCEPT if in an assertion. */
5023            break;  
5024              if (verbs[i].op == OP_ACCEPT)
5025                {
5026                open_capitem *oc;
5027                if (arglen != 0)
5028                  {
5029                  *errorcodeptr = ERR59;
5030                  goto FAILED;
5031                  }
5032                cd->had_accept = TRUE;
5033                for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5034                  {
5035                  *code++ = OP_CLOSE;
5036                  PUT2INC(code, 0, oc->number);
5037                  }
5038                *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5039                }
5040    
5041              /* Handle other cases with/without an argument */
5042    
5043              else if (arglen == 0)
5044                {
5045                if (verbs[i].op < 0)   /* Argument is mandatory */
5046                  {
5047                  *errorcodeptr = ERR66;
5048                  goto FAILED;
5049                  }
5050                *code = verbs[i].op;
5051                if (*code++ == OP_THEN)
5052                  {
5053                  PUT(code, 0, code - bcptr->current_branch - 1);
5054                  code += LINK_SIZE;
5055                  }
5056                }
5057    
5058              else
5059                {
5060                if (verbs[i].op_arg < 0)   /* Argument is forbidden */
5061                  {
5062                  *errorcodeptr = ERR59;
5063                  goto FAILED;
5064                  }
5065                *code = verbs[i].op_arg;
5066                if (*code++ == OP_THEN_ARG)
5067                  {
5068                  PUT(code, 0, code - bcptr->current_branch - 1);
5069                  code += LINK_SIZE;
5070                  }
5071                *code++ = arglen;
5072                memcpy(code, arg, arglen);
5073                code += arglen;
5074                *code++ = 0;
5075                }
5076    
5077              break;  /* Found verb, exit loop */
5078            }            }
5079    
5080          vn += verbs[i].len + 1;          vn += verbs[i].len + 1;
5081          }          }
5082        if (i < verbcount) continue;  
5083        *errorcodeptr = ERR60;        if (i < verbcount) continue;    /* Successfully handled a verb */
5084          *errorcodeptr = ERR60;          /* Verb not recognized */
5085        goto FAILED;        goto FAILED;
5086        }        }
5087    
# Line 4433  we set the flag only if there is a liter Line 5200  we set the flag only if there is a liter
5200                recno * 10 + *ptr - CHAR_0 : -1;                recno * 10 + *ptr - CHAR_0 : -1;
5201            ptr++;            ptr++;
5202            }            }
5203          namelen = ptr - name;          namelen = (int)(ptr - name);
5204    
5205          if ((terminator > 0 && *ptr++ != terminator) ||          if ((terminator > 0 && *ptr++ != terminator) ||
5206              *ptr++ != CHAR_RIGHT_PARENTHESIS)              *ptr++ != CHAR_RIGHT_PARENTHESIS)
# Line 4470  we set the flag only if there is a liter Line 5237  we set the flag only if there is a liter
5237            }            }
5238    
5239          /* Otherwise (did not start with "+" or "-"), start by looking for the          /* Otherwise (did not start with "+" or "-"), start by looking for the
5240          name. */          name. If we find a name, add one to the opcode to change OP_CREF or
5241            OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5242            except they record that the reference was originally to a name. The
5243            information is used to check duplicate names. */
5244    
5245          slot = cd->name_table;          slot = cd->name_table;
5246          for (i = 0; i < cd->names_found; i++)          for (i = 0; i < cd->names_found; i++)
# Line 4485  we set the flag only if there is a liter Line 5255  we set the flag only if there is a liter
5255            {            {
5256            recno = GET2(slot, 0);            recno = GET2(slot, 0);
5257            PUT2(code, 2+LINK_SIZE, recno);            PUT2(code, 2+LINK_SIZE, recno);
5258              code[1+LINK_SIZE]++;
5259            }            }
5260    
5261          /* Search the pattern for a forward reference */          /* Search the pattern for a forward reference */
5262    
5263          else if ((i = find_parens(ptr, cd, name, namelen,          else if ((i = find_parens(cd, name, namelen,
5264                          (options & PCRE_EXTENDED) != 0)) > 0)                          (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5265            {            {
5266            PUT2(code, 2+LINK_SIZE, i);            PUT2(code, 2+LINK_SIZE, i);
5267              code[1+LINK_SIZE]++;
5268            }            }
5269    
5270          /* If terminator == 0 it means that the name followed directly after          /* If terminator == 0 it means that the name followed directly after
# Line 4557  we set the flag only if there is a liter Line 5329  we set the flag only if there is a liter
5329          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5330          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
5331          bravalue = OP_ASSERT;          bravalue = OP_ASSERT;
5332            cd->assert_depth += 1;
5333          ptr++;          ptr++;
5334          break;          break;
5335    
# Line 4571  we set the flag only if there is a liter Line 5344  we set the flag only if there is a liter
5344            continue;            continue;
5345            }            }
5346          bravalue = OP_ASSERT_NOT;          bravalue = OP_ASSERT_NOT;
5347            cd->assert_depth += 1;
5348          break;          break;
5349    
5350    
# Line 4580  we set the flag only if there is a liter Line 5354  we set the flag only if there is a liter
5354            {            {
5355            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
5356            bravalue = OP_ASSERTBACK;            bravalue = OP_ASSERTBACK;
5357              cd->assert_depth += 1;
5358            ptr += 2;            ptr += 2;
5359            break;            break;
5360    
5361            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
5362            bravalue = OP_ASSERTBACK_NOT;            bravalue = OP_ASSERTBACK_NOT;
5363              cd->assert_depth += 1;
5364            ptr += 2;            ptr += 2;
5365            break;            break;
5366    
# Line 4624  we set the flag only if there is a liter Line 5400  we set the flag only if there is a liter
5400              goto FAILED;              goto FAILED;
5401              }              }
5402            *code++ = n;            *code++ = n;
5403            PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */            PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5404            PUT(code, LINK_SIZE, 0);                    /* Default length */            PUT(code, LINK_SIZE, 0);                          /* Default length */
5405            code += 2 * LINK_SIZE;            code += 2 * LINK_SIZE;
5406            }            }
5407          previous = NULL;          previous = NULL;
# Line 4658  we set the flag only if there is a liter Line 5434  we set the flag only if there is a liter
5434            name = ++ptr;            name = ++ptr;
5435    
5436            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5437            namelen = ptr - name;            namelen = (int)(ptr - name);
5438    
5439            /* In the pre-compile phase, just do a syntax check. */            /* In the pre-compile phase, just do a syntax check. */
5440    
# Line 4685  we set the flag only if there is a liter Line 5461  we set the flag only if there is a liter
5461                }                }
5462              }              }
5463    
5464            /* In the real compile, create the entry in the table */            /* In the real compile, create the entry in the table, maintaining
5465              alphabetical order. Duplicate names for different numbers are
5466              permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5467              number are always OK. (An existing number can be re-used if (?|
5468              appears in the pattern.) In either event, a duplicate name results in
5469              a duplicate entry in the table, even if the number is the same. This
5470              is because the number of names, and hence the table size, is computed
5471              in the pre-compile, and it affects various numbers and pointers which
5472              would all have to be modified, and the compiled code moved down, if
5473              duplicates with the same number were omitted from the table. This
5474              doesn't seem worth the hassle. However, *different* names for the
5475              same number are not permitted. */
5476    
5477            else            else
5478              {              {
5479                BOOL dupname = FALSE;
5480              slot = cd->name_table;              slot = cd->name_table;
5481    
5482              for (i = 0; i < cd->names_found; i++)              for (i = 0; i < cd->names_found; i++)
5483                {                {
5484                int crc = memcmp(name, slot+2, namelen);                int crc = memcmp(name, slot+2, namelen);
# Line 4697  we set the flag only if there is a liter Line 5486  we set the flag only if there is a liter
5486                  {                  {
5487                  if (slot[2+namelen] == 0)                  if (slot[2+namelen] == 0)
5488                    {                    {
5489                    if ((options & PCRE_DUPNAMES) == 0)                    if (GET2(slot, 0) != cd->bracount + 1 &&
5490                          (options & PCRE_DUPNAMES) == 0)
5491                      {                      {
5492                      *errorcodeptr = ERR43;                      *errorcodeptr = ERR43;
5493                      goto FAILED;                      goto FAILED;
5494                      }                      }
5495                      else dupname = TRUE;
5496                    }                    }
5497                  else crc = -1;      /* Current name is substring */                  else crc = -1;      /* Current name is a substring */
5498                  }                  }
5499    
5500                  /* Make space in the table and break the loop for an earlier
5501                  name. For a duplicate or later name, carry on. We do this for
5502                  duplicates so that in the simple case (when ?(| is not used) they
5503                  are in order of their numbers. */
5504    
5505                if (crc < 0)                if (crc < 0)
5506                  {                  {
5507                  memmove(slot + cd->name_entry_size, slot,                  memmove(slot + cd->name_entry_size, slot,
5508                    (cd->names_found - i) * cd->name_entry_size);                    (cd->names_found - i) * cd->name_entry_size);
5509                  break;                  break;
5510                  }                  }
5511    
5512                  /* Continue the loop for a later or duplicate name */
5513    
5514                slot += cd->name_entry_size;                slot += cd->name_entry_size;
5515                }                }
5516    
5517                /* For non-duplicate names, check for a duplicate number before
5518                adding the new name. */
5519    
5520                if (!dupname)
5521                  {
5522                  uschar *cslot = cd->name_table;
5523                  for (i = 0; i < cd->names_found; i++)
5524                    {
5525                    if (cslot != slot)
5526                      {
5527                      if (GET2(cslot, 0) == cd->bracount + 1)
5528                        {
5529                        *errorcodeptr = ERR65;
5530                        goto FAILED;
5531                        }
5532                      }
5533                    else i--;
5534                    cslot += cd->name_entry_size;
5535                    }
5536                  }
5537    
5538              PUT2(slot, 0, cd->bracount + 1);              PUT2(slot, 0, cd->bracount + 1);
5539              memcpy(slot + 2, name, namelen);              memcpy(slot + 2, name, namelen);
5540              slot[2+namelen] = 0;              slot[2+namelen] = 0;
5541              }              }
5542            }            }
5543    
5544          /* In both cases, count the number of names we've encountered. */          /* In both pre-compile and compile, count the number of names we've
5545            encountered. */
5546    
         ptr++;                    /* Move past > or ' */  
5547          cd->names_found++;          cd->names_found++;
5548            ptr++;                    /* Move past > or ' */
5549          goto NUMBERED_GROUP;          goto NUMBERED_GROUP;
5550    
5551    
# Line 4742  we set the flag only if there is a liter Line 5564  we set the flag only if there is a liter
5564          NAMED_REF_OR_RECURSE:          NAMED_REF_OR_RECURSE:
5565          name = ++ptr;          name = ++ptr;
5566          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5567          namelen = ptr - name;          namelen = (int)(ptr - name);
5568    
5569          /* 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
5570          reference number. */          a dummy reference number, because it was not used in the first pass.
5571            However, with the change of recursive back references to be atomic,
5572            we have to look for the number so that this state can be identified, as
5573            otherwise the incorrect length is computed. If it's not a backwards
5574            reference, the dummy number will do. */
5575    
5576          if (lengthptr != NULL)          if (lengthptr != NULL)
5577            {            {
5578              const uschar *temp;
5579    
5580            if (namelen == 0)            if (namelen == 0)
5581              {              {
5582              *errorcodeptr = ERR62;              *errorcodeptr = ERR62;
# Line 4764  we set the flag only if there is a liter Line 5592  we set the flag only if there is a liter
5592              *errorcodeptr = ERR48;              *errorcodeptr = ERR48;
5593              goto FAILED;              goto FAILED;
5594              }              }
5595            recno = 0;  
5596              /* The name table does not exist in the first pass, so we cannot
5597              do a simple search as in the code below. Instead, we have to scan the
5598              pattern to find the number. It is important that we scan it only as
5599              far as we have got because the syntax of named subpatterns has not
5600              been checked for the rest of the pattern, and find_parens() assumes
5601              correct syntax. In any case, it's a waste of resources to scan
5602              further. We stop the scan at the current point by temporarily
5603              adjusting the value of cd->endpattern. */
5604    
5605              temp = cd->end_pattern;
5606              cd->end_pattern = ptr;
5607              recno = find_parens(cd, name, namelen,
5608                (options & PCRE_EXTENDED) != 0, utf8);
5609              cd->end_pattern = temp;
5610              if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
5611            }            }
5612    
5613          /* 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 4788  we set the flag only if there is a liter Line 5631  we set the flag only if there is a liter
5631              recno = GET2(slot, 0);              recno = GET2(slot, 0);
5632              }              }
5633            else if ((recno =                /* Forward back reference */            else if ((recno =                /* Forward back reference */
5634                      find_parens(ptr, cd, name, namelen,                      find_parens(cd, name, namelen,
5635                        (options & PCRE_EXTENDED) != 0)) <= 0)                        (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5636              {              {
5637              *errorcodeptr = ERR15;              *errorcodeptr = ERR15;
5638              goto FAILED;              goto FAILED;
# Line 4892  we set the flag only if there is a liter Line 5735  we set the flag only if there is a liter
5735            if (lengthptr == NULL)            if (lengthptr == NULL)
5736              {              {
5737              *code = OP_END;              *code = OP_END;
5738              if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);              if (recno != 0)
5739                  called = _pcre_find_bracket(cd->start_code, utf8, recno);
5740    
5741              /* Forward reference */              /* Forward reference */
5742    
5743              if (called == NULL)              if (called == NULL)
5744                {                {
5745                if (find_parens(ptr, cd, NULL, recno,                if (find_parens(cd, NULL, recno,
5746                      (options & PCRE_EXTENDED) != 0) < 0)                      (options & PCRE_EXTENDED) != 0, utf8) < 0)
5747                  {                  {
5748                  *errorcodeptr = ERR15;                  *errorcodeptr = ERR15;
5749                  goto FAILED;                  goto FAILED;
5750                  }                  }
5751    
5752                  /* Fudge the value of "called" so that when it is inserted as an
5753                  offset below, what it actually inserted is the reference number
5754                  of the group. Then remember the forward reference. */
5755    
5756                called = cd->start_code + recno;                called = cd->start_code + recno;
5757                PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);                PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5758                }                }
5759    
5760              /* If not a forward reference, and the subpattern is still open,              /* If not a forward reference, and the subpattern is still open,
# Line 4913  we set the flag only if there is a liter Line 5762  we set the flag only if there is a liter
5762              recursion that could loop for ever, and diagnose that case. */              recursion that could loop for ever, and diagnose that case. */
5763    
5764              else if (GET(called, 1) == 0 &&              else if (GET(called, 1) == 0 &&
5765                       could_be_empty(called, code, bcptr, utf8))                       could_be_empty(called, code, bcptr, utf8, cd))
5766                {                {
5767                *errorcodeptr = ERR40;                *errorcodeptr = ERR40;
5768                goto FAILED;                goto FAILED;
5769                }                }
5770              }              }
5771    
5772            /* Insert the recursion/subroutine item, automatically wrapped inside            /* Insert the recursion/subroutine item. */
5773            "once" brackets. Set up a "previous group" length so that a  
           subsequent quantifier will work. */  
   
           *code = OP_ONCE;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
   
5774            *code = OP_RECURSE;            *code = OP_RECURSE;
5775            PUT(code, 1, called - cd->start_code);            PUT(code, 1, (int)(called - cd->start_code));
           code += 1 + LINK_SIZE;  
   
           *code = OP_KET;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
5776            code += 1 + LINK_SIZE;            code += 1 + LINK_SIZE;
   
           length_prevgroup = 3 + 3*LINK_SIZE;  
5777            }            }
5778    
5779          /* Can't determine a first byte now */          /* Can't determine a first byte now */
# Line 4997  we set the flag only if there is a liter Line 5834  we set the flag only if there is a liter
5834          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
5835          both phases.          both phases.
5836    
5837          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
5838          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. */  
5839    
5840          if (*ptr == CHAR_RIGHT_PARENTHESIS)          if (*ptr == CHAR_RIGHT_PARENTHESIS)
5841            {            {
# Line 5008  we set the flag only if there is a liter Line 5844  we set the flag only if there is a liter
5844              {              {
5845              cd->external_options = newoptions;              cd->external_options = newoptions;
5846              }              }
5847           else            else
5848              {              {
             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))  
               {  
               *code++ = OP_OPT;  
               *code++ = newoptions & PCRE_IMS;  
               }  
5849              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5850              greedy_non_default = greedy_default ^ 1;              greedy_non_default = greedy_default ^ 1;
5851              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5852              }              }
5853    
5854            /* Change options at this level, and pass them back for use            /* Change options at this level, and pass them back for use
5855            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). */  
5856    
5857            *optionsptr = options = newoptions;            *optionsptr = options = newoptions;
5858            previous = NULL;       /* This item can't be repeated */            previous = NULL;       /* This item can't be repeated */
# Line 5040  we set the flag only if there is a liter Line 5869  we set the flag only if there is a liter
5869          }     /* End of switch for character following (? */          }     /* End of switch for character following (? */
5870        }       /* End of (? handling */        }       /* End of (? handling */
5871    
5872      /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,      /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5873      all unadorned brackets become non-capturing and behave like (?:...)      is set, all unadorned brackets become non-capturing and behave like (?:...)
5874      brackets. */      brackets. */
5875    
5876      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
# Line 5059  we set the flag only if there is a liter Line 5888  we set the flag only if there is a liter
5888        skipbytes = 2;        skipbytes = 2;
5889        }        }
5890    
5891      /* Process nested bracketed regex. Assertions may not be repeated, but      /* Process nested bracketed regex. Assertions used not to be repeatable,
5892      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
5893      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
5894      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. */  
5895    
5896      previous = (bravalue >= OP_ONCE)? code : NULL;      previous = code;                   /* For handling repetition */
5897      *code = bravalue;      *code = bravalue;
5898      tempcode = code;      tempcode = code;
5899      tempreqvary = cd->req_varyopt;     /* Save value before bracket */      tempreqvary = cd->req_varyopt;     /* Save value before bracket */
# Line 5073  we set the flag only if there is a liter Line 5901  we set the flag only if there is a liter
5901    
5902      if (!compile_regex(      if (!compile_regex(
5903           newoptions,                   /* The complete new option state */           newoptions,                   /* The complete new option state */
          options & PCRE_IMS,           /* The previous ims option state */  
5904           &tempcode,                    /* Where to put code (updated) */           &tempcode,                    /* Where to put code (updated) */
5905           &ptr,                         /* Input pointer (updated) */           &ptr,                         /* Input pointer (updated) */
5906           errorcodeptr,                 /* Where to put an error message */           errorcodeptr,                 /* Where to put an error message */
# Line 5089  we set the flag only if there is a liter Line 5916  we set the flag only if there is a liter
5916             &length_prevgroup           /* Pre-compile phase */             &length_prevgroup           /* Pre-compile phase */
5917           ))           ))
5918        goto FAILED;        goto FAILED;
5919    
5920        if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5921          cd->assert_depth -= 1;
5922    
5923      /* 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
5924      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
# Line 5160  we set the flag only if there is a liter Line 5990  we set the flag only if there is a liter
5990          goto FAILED;          goto FAILED;
5991          }          }
5992        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5993        *code++ = OP_BRA;        code++;   /* This already contains bravalue */
5994        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
5995        *code++ = OP_KET;        *code++ = OP_KET;
5996        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
# Line 5233  we set the flag only if there is a liter Line 6063  we set the flag only if there is a liter
6063    
6064      /* ===================================================================*/      /* ===================================================================*/
6065      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6066      are arranged to be the negation of the corresponding OP_values. For the      are arranged to be the negation of the corresponding OP_values in the
6067      back references, the values are ESC_REF plus the reference number. Only      default case when PCRE_UCP is not set. For the back references, the values
6068      back references and those types that consume a character may be repeated.      are ESC_REF plus the reference number. Only back references and those types
6069      We can test for values between ESC_b and ESC_Z for the latter; this may      that consume a character may be repeated. We can test for values between
6070      have to change if any new ones are ever created. */      ESC_b and ESC_Z for the latter; this may have to change if any new ones are
6071        ever created. */
6072    
6073      case CHAR_BACKSLASH:      case CHAR_BACKSLASH:
6074      tempptr = ptr;      tempptr = ptr;
# Line 5327  we set the flag only if there is a liter Line 6158  we set the flag only if there is a liter
6158          }          }
6159    
6160        /* \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).
6161        We also support \k{name} (.NET syntax) */        We also support \k{name} (.NET syntax).  */
6162    
6163        if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||        if (-c == ESC_k)
           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))  
6164          {          {
6165            if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6166              ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6167              {
6168              *errorcodeptr = ERR69;
6169              break;
6170              }
6171          is_recurse = FALSE;          is_recurse = FALSE;
6172          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6173            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6174            CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;            CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6175          goto NAMED_REF_OR_RECURSE;          goto NAMED_REF_OR_RECURSE;
6176          }          }
6177    
6178        /* Back references are handled specially; must disable firstbyte if        /* Back references are handled specially; must disable firstbyte if
6179        not set to cope with cases like (?=(\w+))\1: which would otherwise set        not set to cope with cases like (?=(\w+))\1: which would otherwise set
# Line 5345  we set the flag only if there is a liter Line 6181  we set the flag only if there is a liter
6181    
6182        if (-c >= ESC_REF)        if (-c >= ESC_REF)
6183          {          {
6184            open_capitem *oc;
6185          recno = -c - ESC_REF;          recno = -c - ESC_REF;
6186    
6187          HANDLE_REFERENCE:    /* Come here from named backref handling */          HANDLE_REFERENCE:    /* Come here from named backref handling */
6188          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6189          previous = code;          previous = code;
6190          *code++ = OP_REF;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6191          PUT2INC(code, 0, recno);          PUT2INC(code, 0, recno);
6192          cd->backref_map |= (recno < 32)? (1 << recno) : 1;          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6193          if (recno > cd->top_backref) cd->top_backref = recno;          if (recno > cd->top_backref) cd->top_backref = recno;
6194    
6195            /* Check to see if this back reference is recursive, that it, it
6196            is inside the group that it references. A flag is set so that the
6197            group can be made atomic. */
6198    
6199            for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6200              {
6201              if (oc->number == recno)
6202                {
6203                oc->flag = TRUE;
6204                break;
6205                }
6206              }
6207          }          }
6208    
6209        /* So are Unicode property matches, if supported. */        /* So are Unicode property matches, if supported. */
# Line 5383  we set the flag only if there is a liter Line 6233  we set the flag only if there is a liter
6233  #endif  #endif
6234    
6235        /* For the rest (including \X when Unicode properties are supported), we        /* For the rest (including \X when Unicode properties are supported), we
6236        can obtain the OP value by negating the escape value. */        can obtain the OP value by negating the escape value in the default
6237          situation when PCRE_UCP is not set. When it *is* set, we substitute
6238          Unicode property tests. */
6239    
6240        else        else
6241          {          {
6242          previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;  #ifdef SUPPORT_UCP
6243          *code++ = -c;          if (-c >= ESC_DU && -c <= ESC_wu)
6244              {
6245              nestptr = ptr + 1;                   /* Where to resume */
6246              ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
6247              }
6248            else
6249    #endif
6250              {
6251              previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6252              *code++ = -c;
6253              }
6254          }          }
6255        continue;        continue;
6256        }        }
# Line 5433  we set the flag only if there is a liter Line 6295  we set the flag only if there is a liter
6295    
6296      ONE_CHAR:      ONE_CHAR:
6297      previous = code;      previous = code;
6298      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6299      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6300    
6301      /* Remember if \r or \n were seen */      /* Remember if \r or \n were seen */
# Line 5497  return FALSE; Line 6359  return FALSE;
6359  /* 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
6360  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
6361  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.  
   
6362  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
6363  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
6364  value of lengthptr distinguishes the two phases.  value of lengthptr distinguishes the two phases.
6365    
6366  Arguments:  Arguments:
6367    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  
6368    codeptr        -> the address of the current code pointer    codeptr        -> the address of the current code pointer
6369    ptrptr         -> the address of the current pattern pointer    ptrptr         -> the address of the current pattern pointer
6370    errorcodeptr   -> pointer to error code variable    errorcodeptr   -> pointer to error code variable
# Line 5526  Returns:         TRUE on success Line 6382  Returns:         TRUE on success
6382  */  */
6383    
6384  static BOOL  static BOOL
6385  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,  compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6386    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6387    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6388    int *lengthptr)    int *lengthptr)
# Line 5536  uschar *code = *codeptr; Line 6392  uschar *code = *codeptr;
6392  uschar *last_branch = code;  uschar *last_branch = code;
6393  uschar *start_bracket = code;  uschar *start_bracket = code;
6394  uschar *reverse_count = NULL;  uschar *reverse_count = NULL;
6395    open_capitem capitem;
6396    int capnumber = 0;
6397  int firstbyte, reqbyte;  int firstbyte, reqbyte;
6398  int branchfirstbyte, branchreqbyte;  int branchfirstbyte, branchreqbyte;
6399  int length;  int length;
# Line 5544  int max_bracount; Line 6402  int max_bracount;
6402  branch_chain bc;  branch_chain bc;
6403    
6404  bc.outer = bcptr;  bc.outer = bcptr;
6405  bc.current = code;  bc.current_branch = code;
6406    
6407  firstbyte = reqbyte = REQ_UNSET;  firstbyte = reqbyte = REQ_UNSET;
6408    
# Line 5562  the code that abstracts option settings Line 6420  the code that abstracts option settings
6420  them global. It tests the value of length for (2 + 2*LINK_SIZE) in the  them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6421  pre-compile phase to find out whether anything has yet been compiled or not. */  pre-compile phase to find out whether anything has yet been compiled or not. */
6422    
6423    /* If this is a capturing subpattern, add to the chain of open capturing items
6424    so that we can detect them if (*ACCEPT) is encountered. This is also used to
6425    detect groups that contain recursive back references to themselves. Note that
6426    only OP_CBRA need be tested here; changing this opcode to one of its variants,
6427    e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6428    
6429    if (*code == OP_CBRA)
6430      {
6431      capnumber = GET2(code, 1 + LINK_SIZE);
6432      capitem.number = capnumber;
6433      capitem.next = cd->open_caps;
6434      capitem.flag = FALSE;
6435      cd->open_caps = &capitem;
6436      }
6437    
6438  /* Offset is set zero to mark that this bracket is still open */  /* Offset is set zero to mark that this bracket is still open */
6439    
6440  PUT(code, 1, 0);  PUT(code, 1, 0);
# Line 5577  for (;;) Line 6450  for (;;)
6450    
6451    if (reset_bracount) cd->bracount = orig_bracount;    if (reset_bracount) cd->bracount = orig_bracount;
6452    
   /* 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;  
     }  
   
6453    /* Set up dummy OP_REVERSE if lookbehind assertion */    /* Set up dummy OP_REVERSE if lookbehind assertion */
6454    
6455    if (lookbehind)    if (lookbehind)
# Line 5656  for (;;) Line 6520  for (;;)
6520    
6521      /* If lookbehind, check that this branch matches a fixed-length string, and      /* If lookbehind, check that this branch matches a fixed-length string, and
6522      put the length into the OP_REVERSE item. Temporarily mark the end of the      put the length into the OP_REVERSE item. Temporarily mark the end of the
6523      branch with OP_END. */      branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6524        because there may be forward references that we can't check here. Set a
6525        flag to cause another lookbehind check at the end. Why not do it all at the
6526        end? Because common, erroneous checks are picked up here and the offset of
6527        the problem can be shown. */
6528    
6529      if (lookbehind)      if (lookbehind)
6530        {        {
6531        int fixed_length;        int fixed_length;
6532        *code = OP_END;        *code = OP_END;
6533        fixed_length = find_fixedlength(last_branch, options);        fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
6534            FALSE, cd);
6535        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
6536        if (fixed_length < 0)        if (fixed_length == -3)
6537            {
6538            cd->check_lookbehind = TRUE;
6539            }
6540          else if (fixed_length < 0)
6541          {          {
6542          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6543          *ptrptr = ptr;          *ptrptr = ptr;
6544          return FALSE;          return FALSE;
6545          }          }
6546        PUT(reverse_count, 0, fixed_length);        else { PUT(reverse_count, 0, fixed_length); }
6547        }        }
6548      }      }
6549    
# Line 5679  for (;;) Line 6552  for (;;)
6552    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
6553    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
6554    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
6555    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. */  
6556    
6557    if (*ptr != CHAR_VERTICAL_LINE)    if (*ptr != CHAR_VERTICAL_LINE)
6558      {      {
6559      if (lengthptr == NULL)      if (lengthptr == NULL)
6560        {        {
6561        int branch_length = code - last_branch;        int branch_length = (int)(code - last_branch);
6562        do        do
6563          {          {
6564          int prev_length = GET(last_branch, 1);          int prev_length = GET(last_branch, 1);
# Line 5701  for (;;) Line 6572  for (;;)
6572      /* Fill in the ket */      /* Fill in the ket */
6573    
6574      *code = OP_KET;      *code = OP_KET;
6575      PUT(code, 1, code - start_bracket);      PUT(code, 1, (int)(code - start_bracket));
6576      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6577    
6578      /* Resetting option if needed */      /* If it was a capturing subpattern, check to see if it contained any
6579        recursive back references. If so, we must wrap it in atomic brackets.
6580        In any event, remove the block from the chain. */
6581    
6582      if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)      if (capnumber > 0)
6583        {        {
6584        *code++ = OP_OPT;        if (cd->open_caps->flag)
6585        *code++ = oldims;          {
6586        length += 2;          memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6587              code - start_bracket);
6588            *start_bracket = OP_ONCE;
6589            code += 1 + LINK_SIZE;
6590            PUT(start_bracket, 1, (int)(code - start_bracket));
6591            *code = OP_KET;
6592            PUT(code, 1, (int)(code - start_bracket));
6593            code += 1 + LINK_SIZE;
6594            length += 2 + 2*LINK_SIZE;
6595            }
6596          cd->open_caps = cd->open_caps->next;
6597        }        }
6598    
6599      /* Retain the highest bracket number, in case resetting was used. */      /* Retain the highest bracket number, in case resetting was used. */
# Line 5752  for (;;) Line 6635  for (;;)
6635    else    else
6636      {      {
6637      *code = OP_ALT;      *code = OP_ALT;
6638      PUT(code, 1, code - last_branch);      PUT(code, 1, (int)(code - last_branch));
6639      bc.current = last_branch = code;      bc.current_branch = last_branch = code;
6640      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6641      }      }
6642    
# Line 5772  for (;;) Line 6655  for (;;)
6655  /* 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
6656  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
6657  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
6658  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
6659  counts, since OP_CIRC can match in the middle.  be found, because ^ generates OP_CIRCM in that mode.
6660    
6661  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.
6662  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 5794  of the more common cases more precisely. Line 6677  of the more common cases more precisely.
6677    
6678  Arguments:  Arguments:
6679    code           points to start of expression (the bracket)    code           points to start of expression (the bracket)
   options        points to the options setting  
6680    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
6681                    handles up to substring 31; after that we just have to take                    handles up to substring 31; after that we just have to take
6682                    the less precise approach                    the less precise approach
# Line 5804  Returns:     TRUE or FALSE Line 6686  Returns:     TRUE or FALSE
6686  */  */
6687    
6688  static BOOL  static BOOL
6689  is_anchored(register const uschar *code, int *options, unsigned int bracket_map,  is_anchored(register const uschar *code, unsigned int bracket_map,
6690    unsigned int backref_map)    unsigned int backref_map)
6691  {  {
6692  do {  do {
6693     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6694       options, PCRE_MULTILINE, FALSE);       FALSE);
6695     register int op = *scode;     register int op = *scode;
6696    
6697     /* Non-capturing brackets */     /* Non-capturing brackets */
6698    
6699     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
6700           op == OP_SBRA || op == OP_SBRAPOS)
6701       {       {
6702       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6703       }       }
6704    
6705     /* Capturing brackets */     /* Capturing brackets */
6706    
6707     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
6708                op == OP_SCBRA || op == OP_SCBRAPOS)
6709       {       {
6710       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
6711       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6712       if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;       if (!is_anchored(scode, new_map, backref_map)) return FALSE;
6713       }       }
6714    
6715     /* Other brackets */     /* Other brackets */
6716    
6717     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6718       {       {
6719       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6720       }       }
6721    
6722     /* .* 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 5847  do { Line 6731  do {
6731    
6732     /* Check for explicit anchoring */     /* Check for explicit anchoring */
6733    
6734     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;  
6735     code += GET(code, 1);     code += GET(code, 1);
6736     }     }
6737  while (*code == OP_ALT);   /* Loop for each alternative */  while (*code == OP_ALT);   /* Loop for each alternative */
# Line 5885  is_startline(const uschar *code, unsigne Line 6767  is_startline(const uschar *code, unsigne
6767  {  {
6768  do {  do {
6769     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6770       NULL, 0, FALSE);       FALSE);
6771     register int op = *scode;     register int op = *scode;
6772    
6773     /* 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 5900  do { Line 6782  do {
6782       switch (*scode)       switch (*scode)
6783         {         {
6784         case OP_CREF:         case OP_CREF:
6785           case OP_NCREF:
6786         case OP_RREF:         case OP_RREF:
6787         &n