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

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

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

revision 454 by ph10, Tue Sep 22 09:42:11 2009 UTC revision 654 by ph10, Tue Aug 2 11:00:40 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 *, int *, branch_chain *, compile_data *, int *);
551    
552    
# 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 1017  top-level call starts at the beginning o Line 1110  top-level call starts at the beginning o
1110  start at a parenthesis. It scans along a pattern's text looking for capturing  start at a parenthesis. It scans along a pattern's text looking for capturing
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. We know that if (?P< is  returns when it reaches a given numbered subpattern. Recursion is used to keep
1114  encountered, the name will be terminated by '>' because that is checked in the  track of subpatterns that reset the capturing group numbers - the (?| feature.
1115  first pass. Recursion is used to keep track of subpatterns that reset the  
1116  capturing group numbers - the (?| feature.  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    ptrptr       address of the current character pointer (updated)    ptrptr       address of the current character pointer (updated)
# Line 1028  Arguments: Line 1132  Arguments:
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)    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
# Line 1035  Returns:       the number of the named s Line 1140  Returns:       the number of the named s
1140    
1141  static int  static int
1142  find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,  find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1143    BOOL xmode, int *count)    BOOL xmode, BOOL utf8, int *count)
1144  {  {
1145  uschar *ptr = *ptrptr;  uschar *ptr = *ptrptr;
1146  int start_count = *count;  int start_count = *count;
# Line 1047  dealing with. The very first call may no Line 1152  dealing with. The very first call may no
1152    
1153  if (ptr[0] == CHAR_LEFT_PARENTHESIS)  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1154    {    {
1155    if (ptr[1] == CHAR_QUESTION_MARK &&    /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1156        ptr[2] == CHAR_VERTICAL_LINE)  
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;      ptr += 3;
1174      dup_parens = TRUE;      dup_parens = TRUE;
1175      }      }
1176    
1177    /* Handle a normal, unnamed capturing parenthesis */    /* Handle comments; all characters are allowed until a ket is reached. */
1178    
1179    else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)    else if (ptr[2] == CHAR_NUMBER_SIGN)
1180      {      {
1181      *count += 1;      for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1182      if (name == NULL && *count == lorn) return *count;      goto FAIL_EXIT;
     ptr++;  
1183      }      }
1184    
1185    /* Handle a condition. If it is an assertion, just carry on so that it    /* 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    is processed as normal. If not, skip to the closing parenthesis of the
1187    condition (there can't be any nested parens. */    condition (there can't be any nested parens). */
1188    
1189    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1190      {      {
# Line 1077  if (ptr[0] == CHAR_LEFT_PARENTHESIS) Line 1196  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1196        }        }
1197      }      }
1198    
1199    /* We have either (? or (* and not a condition */    /* Start with (? but not a condition. */
1200    
1201    else    else
1202      {      {
# Line 1100  if (ptr[0] == CHAR_LEFT_PARENTHESIS) Line 1219  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1219        if (name != NULL && lorn == ptr - thisname &&        if (name != NULL && lorn == ptr - thisname &&
1220            strncmp((const char *)name, (const char *)thisname, lorn) == 0)            strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1221          return *count;          return *count;
1222        term++;        term++;
1223        }        }
1224      }      }
1225    }    }
1226    
1227  /* Past any initial parenthesis handling, scan for parentheses or vertical  /* Past any initial parenthesis handling, scan for parentheses or vertical
1228  bars. */  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 != 0; ptr++)  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    
# Line 1146  for (; *ptr != 0; ptr++) Line 1267  for (; *ptr != 0; ptr++)
1267            break;            break;
1268          }          }
1269        else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)        else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1270          {          {
1271          negate_class = TRUE;          negate_class = TRUE;
1272          ptr++;          ptr++;
1273          }          }
1274        else break;        else break;
1275        }        }
1276    
# Line 1182  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        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;      if (*ptr == 0) goto FAIL_EXIT;
1316      continue;      continue;
1317      }      }
# Line 1191  for (; *ptr != 0; ptr++) Line 1320  for (; *ptr != 0; ptr++)
1320    
1321    if (*ptr == CHAR_LEFT_PARENTHESIS)    if (*ptr == CHAR_LEFT_PARENTHESIS)
1322      {      {
1323      int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);      int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1324      if (rc > 0) return rc;      if (rc > 0) return rc;
1325      if (*ptr == 0) goto FAIL_EXIT;      if (*ptr == 0) goto FAIL_EXIT;
1326      }      }
# Line 1199  for (; *ptr != 0; ptr++) Line 1328  for (; *ptr != 0; ptr++)
1328    else if (*ptr == CHAR_RIGHT_PARENTHESIS)    else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1329      {      {
1330      if (dup_parens && *count < hwm_count) *count = hwm_count;      if (dup_parens && *count < hwm_count) *count = hwm_count;
1331      *ptrptr = ptr;      goto FAIL_EXIT;
     return -1;  
1332      }      }
1333    
1334    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
# Line 1238  Arguments: Line 1366  Arguments:
1366    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1367    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1368    xmode        TRUE if we are in /x mode    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  Returns:       the number of the found subpattern, or -1 if not found
1372  */  */
1373    
1374  static int  static int
1375  find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)  find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1376      BOOL utf8)
1377  {  {
1378  uschar *ptr = (uschar *)cd->start_pattern;  uschar *ptr = (uschar *)cd->start_pattern;
1379  int count = 0;  int count = 0;
# Line 1256  matching closing parens. That is why we Line 1386  matching closing parens. That is why we
1386    
1387  for (;;)  for (;;)
1388    {    {
1389    rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);    rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1390    if (rc > 0 || *ptr++ == 0) break;    if (rc > 0 || *ptr++ == 0) break;
1391    }    }
1392    
# Line 1272  return rc; Line 1402  return rc;
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 1315  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 1336  for (;;) Line 1458  for (;;)
1458    
1459  /* Scan a branch 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. The branch is  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.  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  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.  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,  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  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.  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    atend    TRUE if called when the pattern is complete
1474    cd       the "compile data" structure    cd       the "compile data" structure
1475    
1476  Returns:   the fixed length,  Returns:   the fixed length,
1477               or -1 if there is no fixed length,               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               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, BOOL atend, compile_data *cd)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1484  {  {
1485  int length = -1;  int length = -1;
1486    
# Line 1375  for (;;) Line 1497  for (;;)
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, atend, cd);      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 1388  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 1401  for (;;) Line 1529  for (;;)
1529      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1530      branchlength = 0;      branchlength = 0;
1531      break;      break;
1532    
1533      /* A true recursion implies not fixed length, but a subroutine call may      /* 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      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. */      it until the end of the pattern, so return -3. */
1536    
1537      case OP_RECURSE:      case OP_RECURSE:
1538      if (!atend) return -3;      if (!atend) return -3;
1539      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1540      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1541      if (cc > cs && cc < ce) return -1;                /* Recursion */      if (cc > cs && cc < ce) return -1;                /* Recursion */
1542      d = find_fixedlength(cs + 2, options, atend, cd);      d = find_fixedlength(cs + 2, utf8, atend, cd);
1543      if (d < 0) return d;      if (d < 0) return d;
1544      branchlength += d;      branchlength += d;
1545      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1546      break;      break;
1547    
1548      /* Skip over assertive subpatterns */      /* Skip over assertive subpatterns */
1549    
# Line 1430  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 1448  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 && cc[-1] >= 0xc0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       cc += _pcre_utf8_table4[cc[-1] & 0x3f];  
1590  #endif  #endif
1591      break;      break;
1592    
# Line 1465  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 && cc[-1] >= 0xc0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       cc += _pcre_utf8_table4[cc[-1] & 0x3f];  
1601  #endif  #endif
1602      break;      break;
1603    
# Line 1550  for (;;) Line 1681  for (;;)
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, or, if the number is negative, an  capturing bracket with the given number, or, if the number is negative, an
1684  instance of OP_REVERSE for a lookbehind.  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
# Line 1560  Arguments: Line 1693  Arguments:
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 1573  for (;;) Line 1707  for (;;)
1707    the table is zero; the actual length is stored in the compiled code. */    the table is zero; the actual length is stored in the compiled code. */
1708    
1709    if (c == OP_XCLASS) code += GET(code, 1);    if (c == OP_XCLASS) code += GET(code, 1);
1710    
1711    /* Handle recursion */    /* Handle recursion */
1712    
1713    else if (c == OP_REVERSE)    else if (c == OP_REVERSE)
1714      {      {
1715      if (number < 0) return (uschar *)code;      if (number < 0) return (uschar *)code;
1716      code += _pcre_OP_lengths[c];      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 1593  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 1617  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 1631  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 1688  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 1712  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 1726  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 1768  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 1794  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 1804  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 1822  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 1893  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 1934  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 1942  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];      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2198      break;      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        case OP_POSUPTOI:
2206      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];      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 1972  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 2020  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 2034  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 2170  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 2196  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 2246  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 2259  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 2271  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 2285  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 2322  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 2335  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 2372  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 2393  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 2440  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 2456  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 2525  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 2539  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 2602  Arguments: Line 3035  Arguments:
3035    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)    firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3036    reqbyteptr     set to the last literal character required, else < 0    reqbyteptr     set to the last literal character required, else < 0
3037    bcptr          points to current branch chain    bcptr          points to current branch chain
3038      cond_depth     conditional nesting depth
3039    cd             contains pointers to tables etc.    cd             contains pointers to tables etc.
3040    lengthptr      NULL during the real compile phase    lengthptr      NULL during the real compile phase
3041                   points to length accumulator during pre-compile phase                   points to length accumulator during pre-compile phase
# Line 2613  Returns:         TRUE on success Line 3047  Returns:         TRUE on success
3047  static BOOL  static BOOL
3048  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3049    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3050    compile_data *cd, int *lengthptr)    int cond_depth, compile_data *cd, int *lengthptr)
3051  {  {
3052  int repeat_type, op_type;  int repeat_type, op_type;
3053  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
# Line 2622  int greedy_default, greedy_non_default; Line 3056  int greedy_default, greedy_non_default;
3056  int firstbyte, reqbyte;  int firstbyte, reqbyte;
3057  int zeroreqbyte, zerofirstbyte;  int zeroreqbyte, zerofirstbyte;
3058  int req_caseopt, reqvary, tempreqvary;  int req_caseopt, reqvary, tempreqvary;
3059  int options = *optionsptr;  int options = *optionsptr;               /* May change dynamically */
3060  int after_manual_callout = 0;  int after_manual_callout = 0;
3061  int length_prevgroup = 0;  int length_prevgroup = 0;
3062  register int c;  register int c;
# Line 2634  BOOL inescq = FALSE; Line 3068  BOOL inescq = FALSE;
3068  BOOL groupsetfirstbyte = FALSE;  BOOL groupsetfirstbyte = FALSE;
3069  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
3070  const uschar *tempptr;  const uschar *tempptr;
3071    const uschar *nestptr = NULL;
3072  uschar *previous = NULL;  uschar *previous = NULL;
3073  uschar *previous_callout = NULL;  uschar *previous_callout = NULL;
3074  uschar *save_hwm = NULL;  uschar *save_hwm = NULL;
3075  uschar classbits[32];  uschar classbits[32];
3076    
3077    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3078    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3079    dynamically as we process the pattern. */
3080    
3081  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3082  BOOL class_utf8;  BOOL class_utf8;
3083  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8 = (options & PCRE_UTF8) != 0;
# Line 2650  BOOL utf8 = FALSE; Line 3089  BOOL utf8 = FALSE;
3089  uschar *utf8_char = NULL;  uschar *utf8_char = NULL;
3090  #endif  #endif
3091    
3092  #ifdef DEBUG  #ifdef PCRE_DEBUG
3093  if (lengthptr != NULL) DPRINTF((">> start branch\n"));  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3094  #endif  #endif
3095    
# Line 2704  for (;; ptr++) Line 3143  for (;; ptr++)
3143    
3144    c = *ptr;    c = *ptr;
3145    
3146      /* If we are at the end of a nested substitution, revert to the outer level
3147      string. Nesting only happens one level deep. */
3148    
3149      if (c == 0 && nestptr != NULL)
3150        {
3151        ptr = nestptr;
3152        nestptr = NULL;
3153        c = *ptr;
3154        }
3155    
3156    /* 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
3157    previous cycle of this loop. */    previous cycle of this loop. */
3158    
3159    if (lengthptr != NULL)    if (lengthptr != NULL)
3160      {      {
3161  #ifdef DEBUG  #ifdef PCRE_DEBUG
3162      if (code > cd->hwm) cd->hwm = code;                 /* High water info */      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3163  #endif  #endif
3164      if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */      if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3165        {        {
3166        *errorcodeptr = ERR52;        *errorcodeptr = ERR52;
3167        goto FAILED;        goto FAILED;
# Line 2734  for (;; ptr++) Line 3183  for (;; ptr++)
3183        goto FAILED;        goto FAILED;
3184        }        }
3185    
3186      *lengthptr += code - last_code;      *lengthptr += (int)(code - last_code);
3187      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));
3188    
3189      /* 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 2761  for (;; ptr++) Line 3210  for (;; ptr++)
3210    /* 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
3211    reference list. */    reference list. */
3212    
3213    else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3214      {      {
3215      *errorcodeptr = ERR52;      *errorcodeptr = ERR52;
3216      goto FAILED;      goto FAILED;
# Line 2809  for (;; ptr++) Line 3258  for (;; ptr++)
3258      previous_callout = NULL;      previous_callout = NULL;
3259      }      }
3260    
3261    /* In extended mode, skip white space and comments */    /* In extended mode, skip white space and comments. */
3262    
3263    if ((options & PCRE_EXTENDED) != 0)    if ((options & PCRE_EXTENDED) != 0)
3264      {      {
3265      if ((cd->ctypes[c] & ctype_space) != 0) continue;      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3266      if (c == CHAR_NUMBER_SIGN)      if (c == CHAR_NUMBER_SIGN)
3267        {        {
3268        while (*(++ptr) != 0)        ptr++;
3269          while (*ptr != 0)
3270          {          {
3271          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3272            ptr++;
3273    #ifdef SUPPORT_UTF8
3274            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3275    #endif
3276          }          }
3277        if (*ptr != 0) continue;        if (*ptr != 0) continue;
3278    
# Line 2852  for (;; ptr++) Line 3306  for (;; ptr++)
3306          *errorcodeptr = ERR20;          *errorcodeptr = ERR20;
3307          goto FAILED;          goto FAILED;
3308          }          }
3309        *lengthptr += code - last_code;   /* To include callout length */        *lengthptr += (int)(code - last_code);   /* To include callout length */
3310        DPRINTF((">> end branch\n"));        DPRINTF((">> end branch\n"));
3311        }        }
3312      return TRUE;      return TRUE;
# Line 2863  for (;; ptr++) Line 3317  for (;; ptr++)
3317      the setting of any following char as a first character. */      the setting of any following char as a first character. */
3318    
3319      case CHAR_CIRCUMFLEX_ACCENT:      case CHAR_CIRCUMFLEX_ACCENT:
3320        previous = NULL;
3321      if ((options & PCRE_MULTILINE) != 0)      if ((options & PCRE_MULTILINE) != 0)
3322        {        {
3323        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3324          *code++ = OP_CIRCM;
3325        }        }
3326      previous = NULL;      else *code++ = OP_CIRC;
     *code++ = OP_CIRC;  
3327      break;      break;
3328    
3329      case CHAR_DOLLAR_SIGN:      case CHAR_DOLLAR_SIGN:
3330      previous = NULL;      previous = NULL;
3331      *code++ = OP_DOLL;      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3332      break;      break;
3333    
3334      /* 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 3057  for (;; ptr++) Line 3512  for (;; ptr++)
3512            ptr++;            ptr++;
3513            }            }
3514    
3515          posix_class = check_posix_name(ptr, tempptr - ptr);          posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3516          if (posix_class < 0)          if (posix_class < 0)
3517            {            {
3518            *errorcodeptr = ERR30;            *errorcodeptr = ERR30;
# Line 3071  for (;; ptr++) Line 3526  for (;; ptr++)
3526          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3527            posix_class = 0;            posix_class = 0;
3528    
3529          /* 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
3530          because we may be adding and subtracting from it, and we don't want to          different escape sequences that use Unicode properties. */
3531          subtract bits that may be in the main map already. At the end we or the  
3532          result into the bit map that is being built. */  #ifdef SUPPORT_UCP
3533            if ((options & PCRE_UCP) != 0)
3534              {
3535              int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3536              if (posix_substitutes[pc] != NULL)
3537                {
3538                nestptr = tempptr + 1;
3539                ptr = posix_substitutes[pc] - 1;
3540                continue;
3541                }
3542              }
3543    #endif
3544            /* In the non-UCP case, we build the bit map for the POSIX class in a
3545            chunk of local store because we may be adding and subtracting from it,
3546            and we don't want to subtract bits that may be in the main map already.
3547            At the end we or the result into the bit map that is being built. */
3548    
3549          posix_class *= 3;          posix_class *= 3;
3550    
# Line 3118  for (;; ptr++) Line 3588  for (;; ptr++)
3588    
3589        /* Backslash may introduce a single character, or it may introduce one        /* Backslash may introduce a single character, or it may introduce one
3590        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
3591        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
3592        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
3593        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
3594        character in them, so set class_charcount bigger than one. */        are either treated as literal characters (by default), or are faulted if
3595          PCRE_EXTRA is set. */
3596    
3597        if (c == CHAR_BACKSLASH)        if (c == CHAR_BACKSLASH)
3598          {          {
3599          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3600          if (*errorcodeptr != 0) goto FAILED;          if (*errorcodeptr != 0) goto FAILED;
3601    
3602          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 */  
3603          else if (-c == ESC_Q)            /* Handle start of quoted string */          else if (-c == ESC_Q)            /* Handle start of quoted string */
3604            {            {
3605            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
# Line 3147  for (;; ptr++) Line 3616  for (;; ptr++)
3616            register const uschar *cbits = cd->cbits;            register const uschar *cbits = cd->cbits;
3617            class_charcount += 2;     /* Greater than 1 is what matters */            class_charcount += 2;     /* Greater than 1 is what matters */
3618    
3619            /* Save time by not doing this in the pre-compile phase. */            switch (-c)
   
           if (lengthptr == NULL) switch (-c)  
3620              {              {
3621    #ifdef SUPPORT_UCP
3622                case ESC_du:     /* These are the values given for \d etc */
3623                case ESC_DU:     /* when PCRE_UCP is set. We replace the */
3624                case ESC_wu:     /* escape sequence with an appropriate \p */
3625                case ESC_WU:     /* or \P to test Unicode properties instead */
3626                case ESC_su:     /* of the default ASCII testing. */
3627                case ESC_SU:
3628                nestptr = ptr;
3629                ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
3630                class_charcount -= 2;                /* Undo! */
3631                continue;
3632    #endif
3633              case ESC_d:              case ESC_d:
3634              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3635              continue;              continue;
# Line 3169  for (;; ptr++) Line 3648  for (;; ptr++)
3648              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3649              continue;              continue;
3650    
3651                /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3652                if it was previously set by something earlier in the character
3653                class. */
3654    
3655              case ESC_s:              case ESC_s:
3656              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];              classbits[0] |= cbits[cbit_space];
3657              classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= cbits[cbit_space+1] & ~0x08;
3658                for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3659              continue;              continue;
3660    
3661              case ESC_S:              case ESC_S:
# Line 3180  for (;; ptr++) Line 3664  for (;; ptr++)
3664              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
3665              continue;              continue;
3666    
3667              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)  
             {  
3668              SETBIT(classbits, 0x09); /* VT */              SETBIT(classbits, 0x09); /* VT */
3669              SETBIT(classbits, 0x20); /* SPACE */              SETBIT(classbits, 0x20); /* SPACE */
3670              SETBIT(classbits, 0xa0); /* NSBP */              SETBIT(classbits, 0xa0); /* NSBP */
# Line 3217  for (;; ptr++) Line 3688  for (;; ptr++)
3688                }                }
3689  #endif  #endif
3690              continue;              continue;
             }  
3691    
3692            if (-c == ESC_H)              case ESC_H:
             {  
3693              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3694                {                {
3695                int x = 0xff;                int x = 0xff;
# Line 3262  for (;; ptr++) Line 3731  for (;; ptr++)
3731                }                }
3732  #endif  #endif
3733              continue;              continue;
             }  
3734    
3735            if (-c == ESC_v)              case ESC_v:
             {  
3736              SETBIT(classbits, 0x0a); /* LF */              SETBIT(classbits, 0x0a); /* LF */
3737              SETBIT(classbits, 0x0b); /* VT */              SETBIT(classbits, 0x0b); /* VT */
3738              SETBIT(classbits, 0x0c); /* FF */              SETBIT(classbits, 0x0c); /* FF */
# Line 3281  for (;; ptr++) Line 3748  for (;; ptr++)
3748                }                }
3749  #endif  #endif
3750              continue;              continue;
             }  
3751    
3752            if (-c == ESC_V)              case ESC_V:
             {  
3753              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3754                {                {
3755                int x = 0xff;                int x = 0xff;
# Line 3314  for (;; ptr++) Line 3779  for (;; ptr++)
3779                }                }
3780  #endif  #endif
3781              continue;              continue;
             }  
   
           /* We need to deal with \P and \p in both phases. */  
3782    
3783  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3784            if (-c == ESC_p || -c == ESC_P)              case ESC_p:
3785              {              case ESC_P:
3786              BOOL negated;                {
3787              int pdata;                BOOL negated;
3788              int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);                int pdata;
3789              if (ptype < 0) goto FAILED;                int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3790              class_utf8 = TRUE;                if (ptype < 0) goto FAILED;
3791              *class_utf8data++ = ((-c == ESC_p) != negated)?                class_utf8 = TRUE;
3792                XCL_PROP : XCL_NOTPROP;                *class_utf8data++ = ((-c == ESC_p) != negated)?
3793              *class_utf8data++ = ptype;                  XCL_PROP : XCL_NOTPROP;
3794              *class_utf8data++ = pdata;                *class_utf8data++ = ptype;
3795              class_charcount -= 2;   /* Not a < 256 character */                *class_utf8data++ = pdata;
3796              continue;                class_charcount -= 2;   /* Not a < 256 character */
3797              }                continue;
3798                  }
3799  #endif  #endif
3800            /* Unrecognized escapes are faulted if PCRE is running in its              /* Unrecognized escapes are faulted if PCRE is running in its
3801            strict mode. By default, for compatibility with Perl, they are              strict mode. By default, for compatibility with Perl, they are
3802            treated as literals. */              treated as literals. */
3803    
3804            if ((options & PCRE_EXTRA) != 0)              default:
3805              {              if ((options & PCRE_EXTRA) != 0)
3806              *errorcodeptr = ERR7;                {
3807              goto FAILED;                *errorcodeptr = ERR7;
3808                  goto FAILED;
3809                  }
3810                class_charcount -= 2;  /* Undo the default count from above */
3811                c = *ptr;              /* Get the final character and fall through */
3812                break;
3813              }              }
   
           class_charcount -= 2;  /* Undo the default count from above */  
           c = *ptr;              /* Get the final character and fall through */  
3814            }            }
3815    
3816          /* 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 3415  for (;; ptr++) Line 3880  for (;; ptr++)
3880            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3881            if (*errorcodeptr != 0) goto FAILED;            if (*errorcodeptr != 0) goto FAILED;
3882    
3883            /* \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 */  
3884    
3885            if (d < 0)            if (d < 0)
3886              {              {
3887              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  
3888                {                {
3889                ptr = oldptr;                ptr = oldptr;
3890                goto LONE_SINGLE_CHARACTER;  /* A few lines below */                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
# Line 3588  for (;; ptr++) Line 4050  for (;; ptr++)
4050          }          }
4051        }        }
4052    
4053      /* 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.
4054        If we are at the end of an internal nested string, revert to the outer
4055        string. */
4056    
4057        while (((c = *(++ptr)) != 0 ||
4058               (nestptr != NULL &&
4059                 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4060               (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4061    
4062      while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));      /* Check for missing terminating ']' */
4063    
4064      if (c == 0)                          /* Missing terminating ']' */      if (c == 0)
4065        {        {
4066        *errorcodeptr = ERR6;        *errorcodeptr = ERR6;
4067        goto FAILED;        goto FAILED;
4068        }        }
4069    
   
 /* 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  
   
   
4070      /* 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
4071      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
4072      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 3624  we set the flag only if there is a liter Line 4074  we set the flag only if there is a liter
4074    
4075      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
4076      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4077      operate on single-bytes only. This is an historical hangover. Maybe one day      operate on single-bytes characters only. This is an historical hangover.
4078      we can tidy these opcodes to handle multi-byte characters.      Maybe one day we can tidy these opcodes to handle multi-byte characters.
4079    
4080      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
4081      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.
4082      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
4083      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
4084      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
4085      reqbyte, save the previous value for reinstating. */      of reqbyte, save the previous value for reinstating. */
4086    
4087  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4088      if (class_charcount == 1 && !class_utf8 &&      if (class_charcount == 1 && !class_utf8 &&
# Line 3643  we set the flag only if there is a liter Line 4093  we set the flag only if there is a liter
4093        {        {
4094        zeroreqbyte = reqbyte;        zeroreqbyte = reqbyte;
4095    
4096        /* The OP_NOT opcode works on one-byte characters only. */        /* The OP_NOT[I] opcodes work on one-byte characters only. */
4097    
4098        if (negate_class)        if (negate_class)
4099          {          {
4100          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4101          zerofirstbyte = firstbyte;          zerofirstbyte = firstbyte;
4102          *code++ = OP_NOT;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4103          *code++ = class_lastchar;          *code++ = class_lastchar;
4104          break;          break;
4105          }          }
# Line 3680  we set the flag only if there is a liter Line 4130  we set the flag only if there is a liter
4130    
4131      /* If there are characters with values > 255, we have to compile an      /* If there are characters with values > 255, we have to compile an
4132      extended class, with its own opcode, unless there was a negated special      extended class, with its own opcode, unless there was a negated special
4133      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
4134      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
4135      (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
4136      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
4137        actual compiled code. */
4138    
4139  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4140      if (class_utf8 && !should_flip_negation)      if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4141        {        {
4142        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
4143        *code++ = OP_XCLASS;        *code++ = OP_XCLASS;
# Line 3712  we set the flag only if there is a liter Line 4163  we set the flag only if there is a liter
4163        }        }
4164  #endif  #endif
4165    
4166      /* 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
4167      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
4168      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
4169      map into the code vector, negating it if necessary. */      (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4170        negating it if necessary. */
4171    
4172      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4173      if (negate_class)      if (negate_class)
# Line 3775  we set the flag only if there is a liter Line 4227  we set the flag only if there is a liter
4227      op_type = 0;                    /* Default single-char op codes */      op_type = 0;                    /* Default single-char op codes */
4228      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
4229    
4230      /* 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
4231      for an inserted OP_ONCE for the additional '+' extension. */      insert something before it. */
4232    
4233      tempcode = previous;      tempcode = previous;
4234    
# Line 3799  we set the flag only if there is a liter Line 4251  we set the flag only if there is a liter
4251        }        }
4252      else repeat_type = greedy_default;      else repeat_type = greedy_default;
4253    
4254        /* If previous was a recursion call, wrap it in atomic brackets so that
4255        previous becomes the atomic group. All recursions were so wrapped in the
4256        past, but it no longer happens for non-repeated recursions. In fact, the
4257        repeated ones could be re-implemented independently so as not to need this,
4258        but for the moment we rely on the code for repeating groups. */
4259    
4260        if (*previous == OP_RECURSE)
4261          {
4262          memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4263          *previous = OP_ONCE;
4264          PUT(previous, 1, 2 + 2*LINK_SIZE);
4265          previous[2 + 2*LINK_SIZE] = OP_KET;
4266          PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4267          code += 2 + 2 * LINK_SIZE;
4268          length_prevgroup = 3 + 3*LINK_SIZE;
4269    
4270          /* When actually compiling, we need to check whether this was a forward
4271          reference, and if so, adjust the offset. */
4272    
4273          if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4274            {
4275            int offset = GET(cd->hwm, -LINK_SIZE);
4276            if (offset == previous + 1 - cd->start_code)
4277              PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4278            }
4279          }
4280    
4281        /* Now handle repetition for the different types of item. */
4282    
4283      /* If previous was a character match, abolish the item and generate a      /* If previous was a character match, abolish the item and generate a
4284      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
4285      that it is set in reqbyte - it might not be if a sequence such as x{3} is      that it is set in reqbyte - it might not be if a sequence such as x{3} is
4286      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
4287      instead.  */      instead.  */
4288    
4289      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (*previous == OP_CHAR || *previous == OP_CHARI)
4290        {        {
4291          op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4292    
4293        /* 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
4294        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
4295        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 3839  we set the flag only if there is a liter Line 4322  we set the flag only if there is a liter
4322    
4323        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4324            repeat_max < 0 &&            repeat_max < 0 &&
4325            check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
             options, cd))  
4326          {          {
4327          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4328          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3852  we set the flag only if there is a liter Line 4334  we set the flag only if there is a liter
4334      /* If previous was a single negated character ([^a] or similar), we use      /* If previous was a single negated character ([^a] or similar), we use
4335      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-
4336      character repeats by setting opt_type to add a suitable offset into      character repeats by setting opt_type to add a suitable offset into
4337      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
4338      currently used only for single-byte chars. */      are currently used only for single-byte chars. */
4339    
4340      else if (*previous == OP_NOT)      else if (*previous == OP_NOT || *previous == OP_NOTI)
4341        {        {
4342        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4343        c = previous[1];        c = previous[1];
4344        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4345            repeat_max < 0 &&            repeat_max < 0 &&
4346            check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4347          {          {
4348          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4349          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3885  we set the flag only if there is a liter Line 4367  we set the flag only if there is a liter
4367    
4368        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4369            repeat_max < 0 &&            repeat_max < 0 &&
4370            check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4371          {          {
4372          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4373          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3907  we set the flag only if there is a liter Line 4389  we set the flag only if there is a liter
4389    
4390        if (repeat_max == 0) goto END_REPEAT;        if (repeat_max == 0) goto END_REPEAT;
4391    
4392        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4393        /* This code is obsolete from release 8.00; the restriction was finally        /* This code is obsolete from release 8.00; the restriction was finally
4394        removed: */        removed: */
4395    
4396        /* All real repeats make it impossible to handle partial matching (maybe        /* All real repeats make it impossible to handle partial matching (maybe
4397        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4398    
4399        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4400        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4401    
4402        /* Combine the op_type with the repeat_type */        /* Combine the op_type with the repeat_type */
4403    
# Line 4054  we set the flag only if there is a liter Line 4536  we set the flag only if there is a liter
4536  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4537               *previous == OP_XCLASS ||               *previous == OP_XCLASS ||
4538  #endif  #endif
4539               *previous == OP_REF)               *previous == OP_REF ||
4540                 *previous == OP_REFI)
4541        {        {
4542        if (repeat_max == 0)        if (repeat_max == 0)
4543          {          {
# Line 4062  we set the flag only if there is a liter Line 4545  we set the flag only if there is a liter
4545          goto END_REPEAT;          goto END_REPEAT;
4546          }          }
4547    
4548        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4549        /* This code is obsolete from release 8.00; the restriction was finally        /* This code is obsolete from release 8.00; the restriction was finally
4550        removed: */        removed: */
4551    
# Line 4070  we set the flag only if there is a liter Line 4553  we set the flag only if there is a liter
4553        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4554    
4555        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4556        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4557    
4558        if (repeat_min == 0 && repeat_max == -1)        if (repeat_min == 0 && repeat_max == -1)
4559          *code++ = OP_CRSTAR + repeat_type;          *code++ = OP_CRSTAR + repeat_type;
# Line 4088  we set the flag only if there is a liter Line 4571  we set the flag only if there is a liter
4571        }        }
4572    
4573      /* 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
4574      cases. */      cases. Note that at this point we can encounter only the "basic" bracket
4575        opcodes such as BRA and CBRA, as this is the place where they get converted
4576        into the more special varieties such as BRAPOS and SBRA. A test for >=
4577        OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4578        ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4579        repetition of assertions, but now it does, for Perl compatibility. */
4580    
4581      else if (*previous == OP_BRA  || *previous == OP_CBRA ||      else if (*previous >= OP_ASSERT && *previous <= OP_COND)
              *previous == OP_ONCE || *previous == OP_COND)  
4582        {        {
4583        register int i;        register int i;
4584        int ketoffset = 0;        int len = (int)(code - previous);
       int len = code - previous;  
4585        uschar *bralink = NULL;        uschar *bralink = NULL;
4586          uschar *brazeroptr = NULL;
4587    
4588        /* Repeating a DEFINE group is pointless */        /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4589          we just ignore the repeat. */
4590    
4591        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4592          {          goto END_REPEAT;
         *errorcodeptr = ERR55;  
         goto FAILED;  
         }  
4593    
4594        /* If the maximum repeat count is unlimited, find the end of the bracket        /* There is no sense in actually repeating assertions. The only potential
4595        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,
4596        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
4597        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. */
4598        pointer. */  
4599          if (*previous < OP_ONCE)    /* Assertion */
4600        if (repeat_max == -1)          {
4601          {          if (repeat_min > 0) goto END_REPEAT;
4602          register uschar *ket = previous;          if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
         do ket += GET(ket, 1); while (*ket != OP_KET);  
         ketoffset = code - ket;  
4603          }          }
4604    
4605        /* The case of a zero minimum is special because of the need to stick        /* The case of a zero minimum is special because of the need to stick
# Line 4137  we set the flag only if there is a liter Line 4620  we set the flag only if there is a liter
4620          **   goto END_REPEAT;          **   goto END_REPEAT;
4621          **   }          **   }
4622    
4623          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
4624          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
4625          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
4626          groups are referenced, we cannot do this selectively.          don't have a list of which groups are referenced, we cannot do this
4627            selectively.
4628    
4629          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
4630          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 4160  we set the flag only if there is a liter Line 4644  we set the flag only if there is a liter
4644              *previous++ = OP_SKIPZERO;              *previous++ = OP_SKIPZERO;
4645              goto END_REPEAT;              goto END_REPEAT;
4646              }              }
4647              brazeroptr = previous;    /* Save for possessive optimizing */
4648            *previous++ = OP_BRAZERO + repeat_type;            *previous++ = OP_BRAZERO + repeat_type;
4649            }            }
4650    
# Line 4184  we set the flag only if there is a liter Line 4669  we set the flag only if there is a liter
4669            /* We chain together the bracket offset fields that have to be            /* We chain together the bracket offset fields that have to be
4670            filled in later when the ends of the brackets are reached. */            filled in later when the ends of the brackets are reached. */
4671    
4672            offset = (bralink == NULL)? 0 : previous - bralink;            offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4673            bralink = previous;            bralink = previous;
4674            PUTINC(previous, 0, offset);            PUTINC(previous, 0, offset);
4675            }            }
# Line 4205  we set the flag only if there is a liter Line 4690  we set the flag only if there is a liter
4690            {            {
4691            /* 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
4692            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
4693            potential integer overflow. */            potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4694              integer type when available, otherwise double. */
4695    
4696            if (lengthptr != NULL)            if (lengthptr != NULL)
4697              {              {
4698              int delta = (repeat_min - 1)*length_prevgroup;              int delta = (repeat_min - 1)*length_prevgroup;
4699              if ((double)(repeat_min - 1)*(double)length_prevgroup >              if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4700                                                              (double)INT_MAX ||                    (INT64_OR_DOUBLE)length_prevgroup >
4701                        (INT64_OR_DOUBLE)INT_MAX ||
4702                  OFLOW_MAX - *lengthptr < delta)                  OFLOW_MAX - *lengthptr < delta)
4703                {                {
4704                *errorcodeptr = ERR20;                *errorcodeptr = ERR20;
# Line 4257  we set the flag only if there is a liter Line 4744  we set the flag only if there is a liter
4744          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
4745          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
4746          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
4747          paranoid checks to avoid integer overflow. */          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4748            a 64-bit integer type when available, otherwise double. */
4749    
4750          if (lengthptr != NULL && repeat_max > 0)          if (lengthptr != NULL && repeat_max > 0)
4751            {            {
4752            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4753                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */
4754            if ((double)repeat_max *            if ((INT64_OR_DOUBLE)repeat_max *
4755                  (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)                  (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4756                    > (double)INT_MAX ||                    > (INT64_OR_DOUBLE)INT_MAX ||
4757                OFLOW_MAX - *lengthptr < delta)                OFLOW_MAX - *lengthptr < delta)
4758              {              {
4759              *errorcodeptr = ERR20;              *errorcodeptr = ERR20;
# Line 4290  we set the flag only if there is a liter Line 4778  we set the flag only if there is a liter
4778              {              {
4779              int offset;              int offset;
4780              *code++ = OP_BRA;              *code++ = OP_BRA;
4781              offset = (bralink == NULL)? 0 : code - bralink;              offset = (bralink == NULL)? 0 : (int)(code - bralink);
4782              bralink = code;              bralink = code;
4783              PUTINC(code, 0, offset);              PUTINC(code, 0, offset);
4784              }              }
# Line 4311  we set the flag only if there is a liter Line 4799  we set the flag only if there is a liter
4799          while (bralink != NULL)          while (bralink != NULL)
4800            {            {
4801            int oldlinkoffset;            int oldlinkoffset;
4802            int offset = code - bralink + 1;            int offset = (int)(code - bralink + 1);
4803            uschar *bra = code - offset;            uschar *bra = code - offset;
4804            oldlinkoffset = GET(bra, 1);            oldlinkoffset = GET(bra, 1);
4805            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
# Line 4321  we set the flag only if there is a liter Line 4809  we set the flag only if there is a liter
4809            }            }
4810          }          }
4811    
4812        /* 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
4813        can't just offset backwards from the current code point, because we        ONCE brackets, that's all we need to do. However, possessively repeated
4814        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
4815        correct offset was computed above.        behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4816          deal with possessive ONCEs specially.
4817    
4818          Otherwise, if the quantifier was possessive, we convert the BRA code to
4819          the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4820          at runtime to detect this kind of subpattern at both the start and at the
4821          end.) The use of special opcodes makes it possible to reduce greatly the
4822          stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4823          convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4824          the default action below, of wrapping everything inside atomic brackets,
4825          does not happen.
4826    
4827        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
4828        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
4829        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
4830        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
4831        atomic groups at runtime, but in a different way.] */        runtime, but in a different way.] */
4832    
4833        else        else
4834          {          {
4835          uschar *ketcode = code - ketoffset;          uschar *ketcode = code - 1 - LINK_SIZE;
4836          uschar *bracode = ketcode - GET(ketcode, 1);          uschar *bracode = ketcode - GET(ketcode, 1);
4837          *ketcode = OP_KETRMAX + repeat_type;  
4838          if (lengthptr == NULL && *bracode != OP_ONCE)          if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4839            if (*bracode == OP_ONCE)
4840              *ketcode = OP_KETRMAX + repeat_type;
4841            else
4842            {            {
4843            uschar *scode = bracode;            if (possessive_quantifier)
4844            do              {
4845                *bracode += 1;                   /* Switch to xxxPOS opcodes */
4846                *ketcode = OP_KETRPOS;
4847                if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4848                possessive_quantifier = FALSE;
4849                }
4850              else *ketcode = OP_KETRMAX + repeat_type;
4851    
4852              if (lengthptr == NULL)
4853              {              {
4854              if (could_be_empty_branch(scode, ketcode, utf8))              uschar *scode = bracode;
4855                do
4856                {                {
4857                *bracode += OP_SBRA - OP_BRA;                if (could_be_empty_branch(scode, ketcode, utf8, cd))
4858                break;                  {
4859                    *bracode += OP_SBRA - OP_BRA;
4860                    break;
4861                    }
4862                  scode += GET(scode, 1);
4863                }                }
4864              scode += GET(scode, 1);              while (*scode == OP_ALT);
4865              }              }
           while (*scode == OP_ALT);  
4866            }            }
4867          }          }
4868        }        }
# Line 4370  we set the flag only if there is a liter Line 4883  we set the flag only if there is a liter
4883        }        }
4884    
4885      /* If the character following a repeat is '+', or if certain optimization      /* If the character following a repeat is '+', or if certain optimization
4886      tests above succeeded, possessive_quantifier is TRUE. For some of the      tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4887      simpler opcodes, there is an special alternative opcode for this. For      there are special alternative opcodes for this case. For anything else, we
4888      anything else, we wrap the entire repeated item inside OP_ONCE brackets.      wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4889      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
4890      but the special opcodes can optimize it a bit. The repeated item starts at      special opcodes can optimize it.
4891      tempcode, not at previous, which might be the first part of a string whose  
4892      (former) last char we repeated.      Possessively repeated subpatterns have already been handled in the code
4893        just above, so possessive_quantifier is always FALSE for them at this
4894        stage.
4895    
4896        Note that the repeated item starts at tempcode, not at previous, which
4897        might be the first part of a string whose (former) last char we repeated.
4898    
4899      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
4900      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 4385  we set the flag only if there is a liter Line 4903  we set the flag only if there is a liter
4903      if (possessive_quantifier)      if (possessive_quantifier)
4904        {        {
4905        int len;        int len;
4906    
4907        if (*tempcode == OP_TYPEEXACT)        if (*tempcode == OP_TYPEEXACT)
4908          tempcode += _pcre_OP_lengths[*tempcode] +          tempcode += _pcre_OP_lengths[*tempcode] +
4909            ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);            ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4910    
4911        else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)        else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4912          {          {
4913          tempcode += _pcre_OP_lengths[*tempcode];          tempcode += _pcre_OP_lengths[*tempcode];
# Line 4397  we set the flag only if there is a liter Line 4915  we set the flag only if there is a liter
4915          if (utf8 && tempcode[-1] >= 0xc0)          if (utf8 && tempcode[-1] >= 0xc0)
4916            tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];            tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4917  #endif  #endif
4918          }          }
4919    
4920        len = code - tempcode;        len = (int)(code - tempcode);
4921        if (len > 0) switch (*tempcode)        if (len > 0) switch (*tempcode)
4922          {          {
4923          case OP_STAR:  *tempcode = OP_POSSTAR; break;          case OP_STAR:  *tempcode = OP_POSSTAR; break;
# Line 4407  we set the flag only if there is a liter Line 4925  we set the flag only if there is a liter
4925          case OP_QUERY: *tempcode = OP_POSQUERY; break;          case OP_QUERY: *tempcode = OP_POSQUERY; break;
4926          case OP_UPTO:  *tempcode = OP_POSUPTO; break;          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
4927    
4928          case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;          case OP_STARI:  *tempcode = OP_POSSTARI; break;
4929          case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;          case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
4930          case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;          case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4931          case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;          case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
4932    
4933          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
4934          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
4935          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4936          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
4937    
4938            case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
4939            case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
4940            case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4941            case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
4942    
4943            case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
4944            case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
4945            case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4946            case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
4947    
4948            /* Because we are moving code along, we must ensure that any
4949            pending recursive references are updated. */
4950    
4951          default:          default:
4952            *code = OP_END;
4953            adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4954          memmove(tempcode + 1+LINK_SIZE, tempcode, len);          memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4955          code += 1 + LINK_SIZE;          code += 1 + LINK_SIZE;
4956          len += 1 + LINK_SIZE;          len += 1 + LINK_SIZE;
# Line 4453  we set the flag only if there is a liter Line 4986  we set the flag only if there is a liter
4986    
4987      /* First deal with various "verbs" that can be introduced by '*'. */      /* First deal with various "verbs" that can be introduced by '*'. */
4988    
4989      if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)      if (*(++ptr) == CHAR_ASTERISK &&
4990             ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4991        {        {
4992        int i, namelen;        int i, namelen;
4993          int arglen = 0;
4994        const char *vn = verbnames;        const char *vn = verbnames;
4995        const uschar *name = ++ptr;        const uschar *name = ptr + 1;
4996          const uschar *arg = NULL;
4997        previous = NULL;        previous = NULL;
4998        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4999          namelen = (int)(ptr - name);
5000    
5001          /* It appears that Perl allows any characters whatsoever, other than
5002          a closing parenthesis, to appear in arguments, so we no longer insist on
5003          letters, digits, and underscores. */
5004    
5005        if (*ptr == CHAR_COLON)        if (*ptr == CHAR_COLON)
5006          {          {
5007          *errorcodeptr = ERR59;   /* Not supported */          arg = ++ptr;
5008          goto FAILED;          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
5009            arglen = (int)(ptr - arg);
5010          }          }
5011    
5012        if (*ptr != CHAR_RIGHT_PARENTHESIS)        if (*ptr != CHAR_RIGHT_PARENTHESIS)
5013          {          {
5014          *errorcodeptr = ERR60;          *errorcodeptr = ERR60;
5015          goto FAILED;          goto FAILED;
5016          }          }
5017        namelen = ptr - name;  
5018          /* Scan the table of verb names */
5019    
5020        for (i = 0; i < verbcount; i++)        for (i = 0; i < verbcount; i++)
5021          {          {
5022          if (namelen == verbs[i].len &&          if (namelen == verbs[i].len &&
5023              strncmp((char *)name, vn, namelen) == 0)              strncmp((char *)name, vn, namelen) == 0)
5024            {            {
5025            /* Check for open captures before ACCEPT */            /* Check for open captures before ACCEPT and convert it to
5026              ASSERT_ACCEPT if in an assertion. */
5027    
5028            if (verbs[i].op == OP_ACCEPT)            if (verbs[i].op == OP_ACCEPT)
5029              {              {
5030              open_capitem *oc;              open_capitem *oc;
5031              cd->had_accept = TRUE;              if (arglen != 0)
5032                  {
5033                  *errorcodeptr = ERR59;
5034                  goto FAILED;
5035                  }
5036                cd->had_accept = TRUE;
5037              for (oc = cd->open_caps; oc != NULL; oc = oc->next)              for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5038                {                {
5039                *code++ = OP_CLOSE;                *code++ = OP_CLOSE;
5040                PUT2INC(code, 0, oc->number);                PUT2INC(code, 0, oc->number);
5041                }                }
5042              }              *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5043            *code++ = verbs[i].op;              }
5044            break;  
5045              /* Handle other cases with/without an argument */
5046    
5047              else if (arglen == 0)
5048                {
5049                if (verbs[i].op < 0)   /* Argument is mandatory */
5050                  {
5051                  *errorcodeptr = ERR66;
5052                  goto FAILED;
5053                  }
5054                *code = verbs[i].op;
5055                if (*code++ == OP_THEN)
5056                  {
5057                  PUT(code, 0, code - bcptr->current_branch - 1);
5058                  code += LINK_SIZE;
5059                  }
5060                }
5061    
5062              else
5063                {
5064                if (verbs[i].op_arg < 0)   /* Argument is forbidden */
5065                  {
5066                  *errorcodeptr = ERR59;
5067                  goto FAILED;
5068                  }
5069                *code = verbs[i].op_arg;
5070                if (*code++ == OP_THEN_ARG)
5071                  {
5072                  PUT(code, 0, code - bcptr->current_branch - 1);
5073                  code += LINK_SIZE;
5074                  }
5075                *code++ = arglen;
5076                memcpy(code, arg, arglen);
5077                code += arglen;
5078                *code++ = 0;
5079                }
5080    
5081              break;  /* Found verb, exit loop */
5082            }            }
5083    
5084          vn += verbs[i].len + 1;          vn += verbs[i].len + 1;
5085          }          }
5086        if (i < verbcount) continue;  
5087        *errorcodeptr = ERR60;        if (i < verbcount) continue;    /* Successfully handled a verb */
5088          *errorcodeptr = ERR60;          /* Verb not recognized */
5089        goto FAILED;        goto FAILED;
5090        }        }
5091    
# Line 4613  we set the flag only if there is a liter Line 5204  we set the flag only if there is a liter
5204                recno * 10 + *ptr - CHAR_0 : -1;                recno * 10 + *ptr - CHAR_0 : -1;
5205            ptr++;            ptr++;
5206            }            }
5207          namelen = ptr - name;          namelen = (int)(ptr - name);
5208    
5209          if ((terminator > 0 && *ptr++ != terminator) ||          if ((terminator > 0 && *ptr++ != terminator) ||
5210              *ptr++ != CHAR_RIGHT_PARENTHESIS)              *ptr++ != CHAR_RIGHT_PARENTHESIS)
# Line 4650  we set the flag only if there is a liter Line 5241  we set the flag only if there is a liter
5241            }            }
5242    
5243          /* Otherwise (did not start with "+" or "-"), start by looking for the          /* Otherwise (did not start with "+" or "-"), start by looking for the
5244          name. */          name. If we find a name, add one to the opcode to change OP_CREF or
5245            OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5246            except they record that the reference was originally to a name. The
5247            information is used to check duplicate names. */
5248    
5249          slot = cd->name_table;          slot = cd->name_table;
5250          for (i = 0; i < cd->names_found; i++)          for (i = 0; i < cd->names_found; i++)
# Line 4665  we set the flag only if there is a liter Line 5259  we set the flag only if there is a liter
5259            {            {
5260            recno = GET2(slot, 0);            recno = GET2(slot, 0);
5261            PUT2(code, 2+LINK_SIZE, recno);            PUT2(code, 2+LINK_SIZE, recno);
5262              code[1+LINK_SIZE]++;
5263            }            }
5264    
5265          /* Search the pattern for a forward reference */          /* Search the pattern for a forward reference */
5266    
5267          else if ((i = find_parens(cd, name, namelen,          else if ((i = find_parens(cd, name, namelen,
5268                          (options & PCRE_EXTENDED) != 0)) > 0)                          (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5269            {            {
5270            PUT2(code, 2+LINK_SIZE, i);            PUT2(code, 2+LINK_SIZE, i);
5271              code[1+LINK_SIZE]++;
5272            }            }
5273    
5274          /* If terminator == 0 it means that the name followed directly after          /* If terminator == 0 it means that the name followed directly after
# Line 4737  we set the flag only if there is a liter Line 5333  we set the flag only if there is a liter
5333          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5334          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
5335          bravalue = OP_ASSERT;          bravalue = OP_ASSERT;
5336            cd->assert_depth += 1;
5337          ptr++;          ptr++;
5338          break;          break;
5339    
# Line 4751  we set the flag only if there is a liter Line 5348  we set the flag only if there is a liter
5348            continue;            continue;
5349            }            }
5350          bravalue = OP_ASSERT_NOT;          bravalue = OP_ASSERT_NOT;
5351            cd->assert_depth += 1;
5352          break;          break;
5353    
5354    
# Line 4760  we set the flag only if there is a liter Line 5358  we set the flag only if there is a liter
5358            {            {
5359            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
5360            bravalue = OP_ASSERTBACK;            bravalue = OP_ASSERTBACK;
5361              cd->assert_depth += 1;
5362            ptr += 2;            ptr += 2;
5363            break;            break;
5364    
5365            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
5366            bravalue = OP_ASSERTBACK_NOT;            bravalue = OP_ASSERTBACK_NOT;
5367              cd->assert_depth += 1;
5368            ptr += 2;            ptr += 2;
5369            break;            break;
5370    
# Line 4804  we set the flag only if there is a liter Line 5404  we set the flag only if there is a liter
5404              goto FAILED;              goto FAILED;
5405              }              }
5406            *code++ = n;            *code++ = n;
5407            PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */            PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5408            PUT(code, LINK_SIZE, 0);                    /* Default length */            PUT(code, LINK_SIZE, 0);                          /* Default length */
5409            code += 2 * LINK_SIZE;            code += 2 * LINK_SIZE;
5410            }            }
5411          previous = NULL;          previous = NULL;
# Line 4838  we set the flag only if there is a liter Line 5438  we set the flag only if there is a liter
5438            name = ++ptr;            name = ++ptr;
5439    
5440            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5441            namelen = ptr - name;            namelen = (int)(ptr - name);
5442    
5443            /* In the pre-compile phase, just do a syntax check. */            /* In the pre-compile phase, just do a syntax check. */
5444    
# Line 4865  we set the flag only if there is a liter Line 5465  we set the flag only if there is a liter
5465                }                }
5466              }              }
5467    
5468            /* In the real compile, create the entry in the table */            /* In the real compile, create the entry in the table, maintaining
5469              alphabetical order. Duplicate names for different numbers are
5470              permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5471              number are always OK. (An existing number can be re-used if (?|
5472              appears in the pattern.) In either event, a duplicate name results in
5473              a duplicate entry in the table, even if the number is the same. This
5474              is because the number of names, and hence the table size, is computed
5475              in the pre-compile, and it affects various numbers and pointers which
5476              would all have to be modified, and the compiled code moved down, if
5477              duplicates with the same number were omitted from the table. This
5478              doesn't seem worth the hassle. However, *different* names for the
5479              same number are not permitted. */
5480    
5481            else            else
5482              {              {
5483                BOOL dupname = FALSE;
5484              slot = cd->name_table;              slot = cd->name_table;
5485    
5486              for (i = 0; i < cd->names_found; i++)              for (i = 0; i < cd->names_found; i++)
5487                {                {
5488                int crc = memcmp(name, slot+2, namelen);                int crc = memcmp(name, slot+2, namelen);
# Line 4877  we set the flag only if there is a liter Line 5490  we set the flag only if there is a liter
5490                  {                  {
5491                  if (slot[2+namelen] == 0)                  if (slot[2+namelen] == 0)
5492                    {                    {
5493                    if ((options & PCRE_DUPNAMES) == 0)                    if (GET2(slot, 0) != cd->bracount + 1 &&
5494                          (options & PCRE_DUPNAMES) == 0)
5495                      {                      {
5496                      *errorcodeptr = ERR43;                      *errorcodeptr = ERR43;
5497                      goto FAILED;                      goto FAILED;
5498                      }                      }
5499                      else dupname = TRUE;
5500                    }                    }
5501                  else crc = -1;      /* Current name is substring */                  else crc = -1;      /* Current name is a substring */
5502                  }                  }
5503    
5504                  /* Make space in the table and break the loop for an earlier
5505                  name. For a duplicate or later name, carry on. We do this for
5506                  duplicates so that in the simple case (when ?(| is not used) they
5507                  are in order of their numbers. */
5508    
5509                if (crc < 0)                if (crc < 0)
5510                  {                  {
5511                  memmove(slot + cd->name_entry_size, slot,                  memmove(slot + cd->name_entry_size, slot,
5512                    (cd->names_found - i) * cd->name_entry_size);                    (cd->names_found - i) * cd->name_entry_size);
5513                  break;                  break;
5514                  }                  }
5515    
5516                  /* Continue the loop for a later or duplicate name */
5517    
5518                slot += cd->name_entry_size;                slot += cd->name_entry_size;
5519                }                }
5520    
5521                /* For non-duplicate names, check for a duplicate number before
5522                adding the new name. */
5523    
5524                if (!dupname)
5525                  {
5526                  uschar *cslot = cd->name_table;
5527                  for (i = 0; i < cd->names_found; i++)
5528                    {
5529                    if (cslot != slot)
5530                      {
5531                      if (GET2(cslot, 0) == cd->bracount + 1)
5532                        {
5533                        *errorcodeptr = ERR65;
5534                        goto FAILED;
5535                        }
5536                      }
5537                    else i--;
5538                    cslot += cd->name_entry_size;
5539                    }
5540                  }
5541    
5542              PUT2(slot, 0, cd->bracount + 1);              PUT2(slot, 0, cd->bracount + 1);
5543              memcpy(slot + 2, name, namelen);              memcpy(slot + 2, name, namelen);
5544              slot[2+namelen] = 0;              slot[2+namelen] = 0;
5545              }              }
5546            }            }
5547    
5548          /* In both cases, count the number of names we've encountered. */          /* In both pre-compile and compile, count the number of names we've
5549            encountered. */
5550    
         ptr++;                    /* Move past > or ' */  
5551          cd->names_found++;          cd->names_found++;
5552            ptr++;                    /* Move past > or ' */
5553          goto NUMBERED_GROUP;          goto NUMBERED_GROUP;
5554    
5555    
# Line 4922  we set the flag only if there is a liter Line 5568  we set the flag only if there is a liter
5568          NAMED_REF_OR_RECURSE:          NAMED_REF_OR_RECURSE:
5569          name = ++ptr;          name = ++ptr;
5570          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5571          namelen = ptr - name;          namelen = (int)(ptr - name);
5572    
5573          /* 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
5574          reference number. */          a dummy reference number, because it was not used in the first pass.
5575            However, with the change of recursive back references to be atomic,
5576            we have to look for the number so that this state can be identified, as
5577            otherwise the incorrect length is computed. If it's not a backwards
5578            reference, the dummy number will do. */
5579    
5580          if (lengthptr != NULL)          if (lengthptr != NULL)
5581            {            {
5582              const uschar *temp;
5583    
5584            if (namelen == 0)            if (namelen == 0)
5585              {              {
5586              *errorcodeptr = ERR62;              *errorcodeptr = ERR62;
# Line 4944  we set the flag only if there is a liter Line 5596  we set the flag only if there is a liter
5596              *errorcodeptr = ERR48;              *errorcodeptr = ERR48;
5597              goto FAILED;              goto FAILED;
5598              }              }
5599            recno = 0;  
5600              /* The name table does not exist in the first pass, so we cannot
5601              do a simple search as in the code below. Instead, we have to scan the
5602              pattern to find the number. It is important that we scan it only as
5603              far as we have got because the syntax of named subpatterns has not
5604              been checked for the rest of the pattern, and find_parens() assumes
5605              correct syntax. In any case, it's a waste of resources to scan
5606              further. We stop the scan at the current point by temporarily
5607              adjusting the value of cd->endpattern. */
5608    
5609              temp = cd->end_pattern;
5610              cd->end_pattern = ptr;
5611              recno = find_parens(cd, name, namelen,
5612                (options & PCRE_EXTENDED) != 0, utf8);
5613              cd->end_pattern = temp;
5614              if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
5615            }            }
5616    
5617          /* 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 4969  we set the flag only if there is a liter Line 5636  we set the flag only if there is a liter
5636              }              }
5637            else if ((recno =                /* Forward back reference */            else if ((recno =                /* Forward back reference */
5638                      find_parens(cd, name, namelen,                      find_parens(cd, name, namelen,
5639                        (options & PCRE_EXTENDED) != 0)) <= 0)                        (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5640              {              {
5641              *errorcodeptr = ERR15;              *errorcodeptr = ERR15;
5642              goto FAILED;              goto FAILED;
# Line 5072  we set the flag only if there is a liter Line 5739  we set the flag only if there is a liter
5739            if (lengthptr == NULL)            if (lengthptr == NULL)
5740              {              {
5741              *code = OP_END;              *code = OP_END;
5742              if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);              if (recno != 0)
5743                  called = _pcre_find_bracket(cd->start_code, utf8, recno);
5744    
5745              /* Forward reference */              /* Forward reference */
5746    
5747              if (called == NULL)              if (called == NULL)
5748                {                {
5749                if (find_parens(cd, NULL, recno,                if (find_parens(cd, NULL, recno,
5750                      (options & PCRE_EXTENDED) != 0) < 0)                      (options & PCRE_EXTENDED) != 0, utf8) < 0)
5751                  {                  {
5752                  *errorcodeptr = ERR15;                  *errorcodeptr = ERR15;
5753                  goto FAILED;                  goto FAILED;
5754                  }                  }
5755    
5756                  /* Fudge the value of "called" so that when it is inserted as an
5757                  offset below, what it actually inserted is the reference number
5758                  of the group. Then remember the forward reference. */
5759    
5760                called = cd->start_code + recno;                called = cd->start_code + recno;
5761                PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);                PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5762                }                }
5763    
5764              /* If not a forward reference, and the subpattern is still open,              /* If not a forward reference, and the subpattern is still open,
5765              this is a recursive call. We check to see if this is a left              this is a recursive call. We check to see if this is a left
5766              recursion that could loop for ever, and diagnose that case. */              recursion that could loop for ever, and diagnose that case. We
5767                must not, however, do this check if we are in a conditional
5768                subpattern because the condition might be testing for recursion in
5769                a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid.
5770                Forever loops are also detected at runtime, so those that occur in
5771                conditional subpatterns will be picked up then. */
5772    
5773              else if (GET(called, 1) == 0 &&              else if (GET(called, 1) == 0 && cond_depth <= 0 &&
5774                       could_be_empty(called, code, bcptr, utf8))                       could_be_empty(called, code, bcptr, utf8, cd))
5775                {                {
5776                *errorcodeptr = ERR40;                *errorcodeptr = ERR40;
5777                goto FAILED;                goto FAILED;
5778                }                }
5779              }              }
5780    
5781            /* Insert the recursion/subroutine item, automatically wrapped inside            /* Insert the recursion/subroutine item. */
           "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;  
5782    
5783            *code = OP_RECURSE;            *code = OP_RECURSE;
5784            PUT(code, 1, called - cd->start_code);            PUT(code, 1, (int)(called - cd->start_code));
5785            code += 1 + LINK_SIZE;            code += 1 + LINK_SIZE;
   
           *code = OP_KET;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
           code += 1 + LINK_SIZE;  
   
           length_prevgroup = 3 + 3*LINK_SIZE;  
5786            }            }
5787    
5788          /* Can't determine a first byte now */          /* Can't determine a first byte now */
# Line 5177  we set the flag only if there is a liter Line 5843  we set the flag only if there is a liter
5843          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
5844          both phases.          both phases.
5845    
5846          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
5847          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. */  
5848    
5849          if (*ptr == CHAR_RIGHT_PARENTHESIS)          if (*ptr == CHAR_RIGHT_PARENTHESIS)
5850            {            {
# Line 5188  we set the flag only if there is a liter Line 5853  we set the flag only if there is a liter
5853              {              {
5854              cd->external_options = newoptions;              cd->external_options = newoptions;
5855              }              }
5856           else            else
5857              {              {
             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))  
               {  
               *code++ = OP_OPT;  
               *code++ = newoptions & PCRE_IMS;  
               }  
5858              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5859              greedy_non_default = greedy_default ^ 1;              greedy_non_default = greedy_default ^ 1;
5860              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5861              }              }
5862    
5863            /* Change options at this level, and pass them back for use            /* Change options at this level, and pass them back for use
5864            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). */  
5865    
5866            *optionsptr = options = newoptions;            *optionsptr = options = newoptions;
5867            previous = NULL;       /* This item can't be repeated */            previous = NULL;       /* This item can't be repeated */
# Line 5220  we set the flag only if there is a liter Line 5878  we set the flag only if there is a liter
5878          }     /* End of switch for character following (? */          }     /* End of switch for character following (? */
5879        }       /* End of (? handling */        }       /* End of (? handling */
5880    
5881      /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,      /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5882      all unadorned brackets become non-capturing and behave like (?:...)      is set, all unadorned brackets become non-capturing and behave like (?:...)
5883      brackets. */      brackets. */
5884    
5885      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
# Line 5239  we set the flag only if there is a liter Line 5897  we set the flag only if there is a liter
5897        skipbytes = 2;        skipbytes = 2;
5898        }        }
5899    
5900      /* Process nested bracketed regex. Assertions may not be repeated, but      /* Process nested bracketed regex. Assertions used not to be repeatable,
5901      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
5902      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
5903      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. */  
5904    
5905      previous = (bravalue >= OP_ONCE)? code : NULL;      previous = code;                      /* For handling repetition */
5906      *code = bravalue;      *code = bravalue;
5907      tempcode = code;      tempcode = code;
5908      tempreqvary = cd->req_varyopt;     /* Save value before bracket */      tempreqvary = cd->req_varyopt;        /* Save value before bracket */
5909      length_prevgroup = 0;              /* Initialize for pre-compile phase */      length_prevgroup = 0;                 /* Initialize for pre-compile phase */
5910    
5911      if (!compile_regex(      if (!compile_regex(
5912           newoptions,                   /* The complete new option state */           newoptions,                      /* The complete new option state */
5913           options & PCRE_IMS,           /* The previous ims option state */           &tempcode,                       /* Where to put code (updated) */
5914           &tempcode,                    /* Where to put code (updated) */           &ptr,                            /* Input pointer (updated) */
5915           &ptr,                         /* Input pointer (updated) */           errorcodeptr,                    /* Where to put an error message */
          errorcodeptr,                 /* Where to put an error message */  
5916           (bravalue == OP_ASSERTBACK ||           (bravalue == OP_ASSERTBACK ||
5917            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */            bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5918           reset_bracount,               /* True if (?| group */           reset_bracount,                  /* True if (?| group */
5919           skipbytes,                    /* Skip over bracket number */           skipbytes,                       /* Skip over bracket number */
5920           &subfirstbyte,                /* For possible first char */           cond_depth +
5921           &subreqbyte,                  /* For possible last char */             ((bravalue == OP_COND)?1:0),   /* Depth of condition subpatterns */
5922           bcptr,                        /* Current branch chain */           &subfirstbyte,                   /* For possible first char */
5923           cd,                           /* Tables block */           &subreqbyte,                     /* For possible last char */
5924           (lengthptr == NULL)? NULL :   /* Actual compile phase */           bcptr,                           /* Current branch chain */
5925             &length_prevgroup           /* Pre-compile phase */           cd,                              /* Tables block */
5926             (lengthptr == NULL)? NULL :      /* Actual compile phase */
5927               &length_prevgroup              /* Pre-compile phase */
5928           ))           ))
5929        goto FAILED;        goto FAILED;
5930    
5931        if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5932          cd->assert_depth -= 1;
5933    
5934      /* 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
5935      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
5936      and any option resetting that may follow it. The pattern pointer (ptr)      and any option resetting that may follow it. The pattern pointer (ptr)
# Line 5340  we set the flag only if there is a liter Line 6001  we set the flag only if there is a liter
6001          goto FAILED;          goto FAILED;
6002          }          }
6003        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
6004        *code++ = OP_BRA;        code++;   /* This already contains bravalue */
6005        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
6006        *code++ = OP_KET;        *code++ = OP_KET;
6007        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
# Line 5413  we set the flag only if there is a liter Line 6074  we set the flag only if there is a liter
6074    
6075      /* ===================================================================*/      /* ===================================================================*/
6076      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6077      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
6078      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
6079      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
6080      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
6081      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
6082        ever created. */
6083    
6084      case CHAR_BACKSLASH:      case CHAR_BACKSLASH:
6085      tempptr = ptr;      tempptr = ptr;
# Line 5507  we set the flag only if there is a liter Line 6169  we set the flag only if there is a liter
6169          }          }
6170    
6171        /* \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).
6172        We also support \k{name} (.NET syntax) */        We also support \k{name} (.NET syntax).  */
6173    
6174        if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||        if (-c == ESC_k)
           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))  
6175          {          {
6176            if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6177              ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6178              {
6179              *errorcodeptr = ERR69;
6180              break;
6181              }
6182          is_recurse = FALSE;          is_recurse = FALSE;
6183          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6184            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
# Line 5525  we set the flag only if there is a liter Line 6192  we set the flag only if there is a liter
6192    
6193        if (-c >= ESC_REF)        if (-c >= ESC_REF)
6194          {          {
6195            open_capitem *oc;
6196          recno = -c - ESC_REF;          recno = -c - ESC_REF;
6197    
6198          HANDLE_REFERENCE:    /* Come here from named backref handling */          HANDLE_REFERENCE:    /* Come here from named backref handling */
6199          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6200          previous = code;          previous = code;
6201          *code++ = OP_REF;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6202          PUT2INC(code, 0, recno);          PUT2INC(code, 0, recno);
6203          cd->backref_map |= (recno < 32)? (1 << recno) : 1;          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6204          if (recno > cd->top_backref) cd->top_backref = recno;          if (recno > cd->top_backref) cd->top_backref = recno;
6205    
6206            /* Check to see if this back reference is recursive, that it, it
6207            is inside the group that it references. A flag is set so that the
6208            group can be made atomic. */
6209    
6210            for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6211              {
6212              if (oc->number == recno)
6213                {
6214                oc->flag = TRUE;
6215                break;
6216                }
6217              }
6218          }          }
6219    
6220        /* So are Unicode property matches, if supported. */        /* So are Unicode property matches, if supported. */
# Line 5563  we set the flag only if there is a liter Line 6244  we set the flag only if there is a liter
6244  #endif  #endif
6245    
6246        /* For the rest (including \X when Unicode properties are supported), we        /* For the rest (including \X when Unicode properties are supported), we
6247        can obtain the OP value by negating the escape value. */        can obtain the OP value by negating the escape value in the default
6248          situation when PCRE_UCP is not set. When it *is* set, we substitute
6249          Unicode property tests. */
6250    
6251        else        else
6252          {          {
6253          previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;  #ifdef SUPPORT_UCP
6254          *code++ = -c;          if (-c >= ESC_DU && -c <= ESC_wu)
6255              {
6256              nestptr = ptr + 1;                   /* Where to resume */
6257              ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
6258              }
6259            else
6260    #endif
6261              {
6262              previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6263              *code++ = -c;
6264              }
6265          }          }
6266        continue;        continue;
6267        }        }
# Line 5613  we set the flag only if there is a liter Line 6306  we set the flag only if there is a liter
6306    
6307      ONE_CHAR:      ONE_CHAR:
6308      previous = code;      previous = code;
6309      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6310      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6311    
6312      /* Remember if \r or \n were seen */      /* Remember if \r or \n were seen */
# Line 5677  return FALSE; Line 6370  return FALSE;
6370  /* 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
6371  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
6372  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.  
   
6373  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
6374  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
6375  value of lengthptr distinguishes the two phases.  value of lengthptr distinguishes the two phases.
6376    
6377  Arguments:  Arguments:
6378    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  
6379    codeptr        -> the address of the current code pointer    codeptr        -> the address of the current code pointer
6380    ptrptr         -> the address of the current pattern pointer    ptrptr         -> the address of the current pattern pointer
6381    errorcodeptr   -> pointer to error code variable    errorcodeptr   -> pointer to error code variable
6382    lookbehind     TRUE if this is a lookbehind assertion    lookbehind     TRUE if this is a lookbehind assertion
6383    reset_bracount TRUE to reset the count for each branch    reset_bracount TRUE to reset the count for each branch
6384    skipbytes      skip this many bytes at start (for brackets and OP_COND)    skipbytes      skip this many bytes at start (for brackets and OP_COND)
6385      cond_depth     depth of nesting for conditional subpatterns
6386    firstbyteptr   place to put the first required character, or a negative number    firstbyteptr   place to put the first required character, or a negative number
6387    reqbyteptr     place to put the last required character, or a negative number    reqbyteptr     place to put the last required character, or a negative number
6388    bcptr          pointer to the chain of currently open branches    bcptr          pointer to the chain of currently open branches
# Line 5706  Returns:         TRUE on success Line 6394  Returns:         TRUE on success
6394  */  */
6395    
6396  static BOOL  static BOOL
6397  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,  compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6398    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6399    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,    int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
6400    int *lengthptr)    compile_data *cd, int *lengthptr)
6401  {  {
6402  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
6403  uschar *code = *codeptr;  uschar *code = *codeptr;
# Line 5726  int max_bracount; Line 6414  int max_bracount;
6414  branch_chain bc;  branch_chain bc;
6415    
6416  bc.outer = bcptr;  bc.outer = bcptr;
6417  bc.current = code;  bc.current_branch = code;
6418    
6419  firstbyte = reqbyte = REQ_UNSET;  firstbyte = reqbyte = REQ_UNSET;
6420    
# Line 5745  them global. It tests the value of lengt Line 6433  them global. It tests the value of lengt
6433  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. */
6434    
6435  /* If this is a capturing subpattern, add to the chain of open capturing items  /* If this is a capturing subpattern, add to the chain of open capturing items
6436  so that we can detect them if (*ACCEPT) is encountered. */  so that we can detect them if (*ACCEPT) is encountered. This is also used to
6437    detect groups that contain recursive back references to themselves. Note that
6438    only OP_CBRA need be tested here; changing this opcode to one of its variants,
6439    e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6440    
6441  if (*code == OP_CBRA)  if (*code == OP_CBRA)
6442    {    {
6443    capnumber = GET2(code, 1 + LINK_SIZE);    capnumber = GET2(code, 1 + LINK_SIZE);
6444    capitem.number = capnumber;    capitem.number = capnumber;
6445    capitem.next = cd->open_caps;    capitem.next = cd->open_caps;
6446    cd->open_caps = &capitem;    capitem.flag = FALSE;
6447    }    cd->open_caps = &capitem;
6448      }
6449    
6450  /* Offset is set zero to mark that this bracket is still open */  /* Offset is set zero to mark that this bracket is still open */
6451    
# Line 5770  for (;;) Line 6462  for (;;)
6462    
6463    if (reset_bracount) cd->bracount = orig_bracount;    if (reset_bracount) cd->bracount = orig_bracount;
6464    
   /* 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;  
     }  
   
6465    /* Set up dummy OP_REVERSE if lookbehind assertion */    /* Set up dummy OP_REVERSE if lookbehind assertion */
6466    
6467    if (lookbehind)    if (lookbehind)
# Line 5793  for (;;) Line 6476  for (;;)
6476    into the length. */    into the length. */
6477    
6478    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,    if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6479          &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))          &branchreqbyte, &bc, cond_depth, cd,
6480            (lengthptr == NULL)? NULL : &length))
6481      {      {
6482      *ptrptr = ptr;      *ptrptr = ptr;
6483      return FALSE;      return FALSE;
# Line 5849  for (;;) Line 6533  for (;;)
6533    
6534      /* If lookbehind, check that this branch matches a fixed-length string, and      /* If lookbehind, check that this branch matches a fixed-length string, and
6535      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
6536      branch with OP_END. If the branch contains OP_RECURSE, the result is -3      branch with OP_END. If the branch contains OP_RECURSE, the result is -3
6537      because there may be forward references that we can't check here. Set a      because there may be forward references that we can't check here. Set a
6538      flag to cause another lookbehind check at the end. Why not do it all at the      flag to cause another lookbehind check at the end. Why not do it all at the
6539      end? Because common, erroneous checks are picked up here and the offset of      end? Because common, erroneous checks are picked up here and the offset of
6540      the problem can be shown. */      the problem can be shown. */
6541    
6542      if (lookbehind)      if (lookbehind)
6543        {        {
6544        int fixed_length;        int fixed_length;
6545        *code = OP_END;        *code = OP_END;
6546        fixed_length = find_fixedlength(last_branch, options, FALSE, cd);        fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
6547            FALSE, cd);
6548        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
6549        if (fixed_length == -3)        if (fixed_length == -3)
6550          {          {
6551          cd->check_lookbehind = TRUE;          cd->check_lookbehind = TRUE;
6552          }          }
6553        else if (fixed_length < 0)        else if (fixed_length < 0)
6554          {          {
6555          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
# Line 5880  for (;;) Line 6565  for (;;)
6565    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
6566    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
6567    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
6568    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. */  
6569    
6570    if (*ptr != CHAR_VERTICAL_LINE)    if (*ptr != CHAR_VERTICAL_LINE)
6571      {      {
6572      if (lengthptr == NULL)      if (lengthptr == NULL)
6573        {        {
6574        int branch_length = code - last_branch;        int branch_length = (int)(code - last_branch);
6575        do        do
6576          {          {
6577          int prev_length = GET(last_branch, 1);          int prev_length = GET(last_branch, 1);
# Line 5898  for (;;) Line 6581  for (;;)
6581          }          }
6582        while (branch_length > 0);        while (branch_length > 0);
6583        }        }
   
     /* If it was a capturing subpattern, remove it from the chain. */  
   
     if (capnumber > 0) cd->open_caps = cd->open_caps->next;  
6584    
6585      /* Fill in the ket */      /* Fill in the ket */
6586    
6587      *code = OP_KET;      *code = OP_KET;
6588      PUT(code, 1, code - start_bracket);      PUT(code, 1, (int)(code - start_bracket));
6589      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6590    
6591      /* Resetting option if needed */      /* If it was a capturing subpattern, check to see if it contained any
6592        recursive back references. If so, we must wrap it in atomic brackets.
6593        In any event, remove the block from the chain. */
6594    
6595      if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)      if (capnumber > 0)
6596        {        {
6597        *code++ = OP_OPT;        if (cd->open_caps->flag)
6598        *code++ = oldims;          {
6599        length += 2;          memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6600              code - start_bracket);
6601            *start_bracket = OP_ONCE;
6602            code += 1 + LINK_SIZE;
6603            PUT(start_bracket, 1, (int)(code - start_bracket));
6604            *code = OP_KET;
6605            PUT(code, 1, (int)(code - start_bracket));
6606            code += 1 + LINK_SIZE;
6607            length += 2 + 2*LINK_SIZE;
6608            }
6609          cd->open_caps = cd->open_caps->next;
6610        }        }
6611    
6612      /* Retain the highest bracket number, in case resetting was used. */      /* Retain the highest bracket number, in case resetting was used. */
# Line 5957  for (;;) Line 6648  for (;;)
6648    else    else
6649      {      {
6650      *code = OP_ALT;      *code = OP_ALT;
6651      PUT(code, 1, code - last_branch);      PUT(code, 1, (int)(code - last_branch));
6652      bc.current = last_branch = code;      bc.current_branch = last_branch = code;
6653      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6654      }      }
6655    
# Line 5977  for (;;) Line 6668  for (;;)
6668  /* 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
6669  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
6670  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
6671  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
6672  counts, since OP_CIRC can match in the middle.  be found, because ^ generates OP_CIRCM in that mode.
6673    
6674  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.
6675  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 5999  of the more common cases more precisely. Line 6690  of the more common cases more precisely.
6690    
6691  Arguments:  Arguments:
6692    code           points to start of expression (the bracket)    code           points to start of expression (the bracket)
   options        points to the options setting  
6693    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
6694                    handles up to substring 31; after that we just have to take                    handles up to substring 31; after that we just have to take
6695                    the less precise approach                    the less precise approach
# Line 6009  Returns:     TRUE or FALSE Line 6699  Returns:     TRUE or FALSE
6699  */  */
6700    
6701  static BOOL  static BOOL
6702  is_anchored(register const uschar *code, int *options, unsigned int bracket_map,  is_anchored(register const uschar *code, unsigned int bracket_map,
6703    unsigned int backref_map)    unsigned int backref_map)
6704  {  {
6705  do {  do {
6706     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6707       options, PCRE_MULTILINE, FALSE);       FALSE);
6708     register int op = *scode;     register int op = *scode;
6709    
6710     /* Non-capturing brackets */     /* Non-capturing brackets */
6711    
6712     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
6713           op == OP_SBRA || op == OP_SBRAPOS)
6714       {       {
6715       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6716       }       }
6717    
6718     /* Capturing brackets */     /* Capturing brackets */
6719    
6720     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
6721                op == OP_SCBRA || op == OP_SCBRAPOS)
6722       {       {
6723       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
6724       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6725       if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;       if (!is_anchored(scode, new_map, backref_map)) return FALSE;
6726       }       }
6727    
6728     /* Other brackets */     /* Other brackets */
6729    
6730     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6731       {       {
6732       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6733       }       }
6734    
6735     /* .* 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 6052  do { Line 6744  do {
6744    
6745     /* Check for explicit anchoring */     /* Check for explicit anchoring */
6746    
6747     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;  
6748     code += GET(code, 1);     code += GET(code, 1);
6749     }     }
6750  while (*code == OP_ALT);   /* Loop for each alternative */  while (*code == OP_ALT);   /* Loop for each alternative */
# Line 6090  is_startline(const uschar *code, unsigne Line 6780  is_startline(const uschar *code, unsigne
6780  {  {
6781  do {  do {
6782     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6783       NULL, 0, FALSE);       FALSE);
6784     register int op = *scode;     register int op = *scode;
6785    
6786     /* 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 6105  do { Line 6795  do {
6795       switch (*scode)       switch (*scode)
6796         {         {
6797         case OP_CREF:         case OP_CREF:
6798           case OP_NCREF:
6799         case OP_RREF:         case OP_RREF:
6800           case OP_NRREF:
6801         case OP_DEF:         case OP_DEF:
6802         return FALSE;         return FALSE;
6803    
# Line 6115  do { Line 6807  do {
6807         scode += 1 + LINK_SIZE;         scode += 1 + LINK_SIZE;
6808         break;         break;
6809         }         }
6810       scode = first_significant_code(scode, NULL, 0, FALSE);       scode = first_significant_code(scode, FALSE);
6811       op = *scode;       op = *scode;
6812       }       }
6813    
6814     /* Non-capturing brackets */     /* Non-capturing brackets */
6815    
6816     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
6817           op == OP_SBRA || op == OP_SBRAPOS)
6818       {       {
6819       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6820       }       }
6821    
6822     /* Capturing brackets */     /* Capturing brackets */
6823    
6824     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
6825                op == OP_SCBRA || op == OP_SCBRAPOS)
6826       {       {
6827       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
6828       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
# Line 6152  do { Line 6846  do {
6846    
6847     /* Check for explicit circumflex */     /* Check for explicit circumflex */
6848    
6849     else if (op != OP_CIRC) return FALSE;     else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
6850    
6851     /* Move on to the next alternative */     /* Move on to the next alternative */
6852    
# Line 6178  we return that char, otherwise -1. Line 6872  we return that char, otherwise -1.
6872    
6873  Arguments:  Arguments:
6874    code       points to start of expression (the bracket)    code       points to start of expression (the bracket)
   options    pointer to the options (used to check casing changes)  
6875    inassert   TRUE if in an assertion    inassert   TRUE if in an assertion
6876    
6877  Returns:     -1 or the fixed first char  Returns:     -1 or the fixed first char
6878  */  */
6879    
6880  static int  static int
6881  find_firstassertedchar(const uschar *code, int *options, BOOL inassert)  find_firstassertedchar(const uschar *code, BOOL inassert)
6882  {  {
6883  register int c = -1;  register int c = -1;
6884  do {  do {
6885     int d;     int d;
6886     const uschar *scode =     int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
6887       first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);               *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? 2:0;
6888       const uschar *scode = first_significant_code(code + 1+LINK_SIZE + xl, TRUE);
6889     register int op = *scode;     register int op = *scode;
6890    
6891     switch(op)     switch(op)
# Line 6200  do { Line 6894  do {
6894       return -1;       return -1;
6895    
6896       case OP_BRA:       case OP_BRA:
6897         case OP_BRAPOS:
6898       case OP_CBRA:       case OP_CBRA:
6899         case OP_SCBRA:
6900         case OP_CBRAPOS:
6901         case OP_SCBRAPOS:
6902       case OP_ASSERT:       case OP_ASSERT:
6903       case OP_ONCE:       case OP_ONCE:
6904       case OP_COND:       case OP_COND:
6905       if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)       if ((d = find_firstassertedchar(scode, op == OP_ASSERT)) < 0)