/[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 455 by ph10, Sat Sep 26 19:12:32 2009 UTC revision 637 by ph10, Sun Jul 24 17:44:12 2011 UTC
# Line 6  Line 6 
6  and semantics are as close as possible to those of the Perl 5 language.  and semantics are as close as possible to those of the Perl 5 language.
7    
8                         Written by Philip Hazel                         Written by Philip Hazel
9             Copyright (c) 1997-2009 University of Cambridge             Copyright (c) 1997-2011 University of Cambridge
10    
11  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
12  Redistribution and use in source and binary forms, with or without  Redistribution and use in source and binary forms, with or without
# Line 53  supporting internal functions that are n Line 53  supporting internal functions that are n
53  #include "pcre_internal.h"  #include "pcre_internal.h"
54    
55    
56  /* When DEBUG is defined, we need the pcre_printint() function, which is also  /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57  used by pcretest. DEBUG is not defined when building a production library. */  also used by pcretest. PCRE_DEBUG is not defined when building a production
58    library. */
59    
60  #ifdef DEBUG  #ifdef PCRE_DEBUG
61  #include "pcre_printint.src"  #include "pcre_printint.src"
62  #endif  #endif
63    
# Line 91  is 4 there is plenty of room. */ Line 92  is 4 there is plenty of room. */
92    
93  #define COMPILE_WORK_SIZE (4096)  #define COMPILE_WORK_SIZE (4096)
94    
95    /* The overrun tests check for a slightly smaller size so that they detect the
96    overrun before it actually does run off the end of the data block. */
97    
98    #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99    
100    
101  /* Table for handling escaped characters in the range '0'-'z'. Positive returns  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102  are simple data values; negative values are for special things like \d and so  are simple data values; negative values are for special things like \d and so
# Line 118  static const short int escapes[] = { Line 124  static const short int escapes[] = {
124       -ESC_H,                  0,       -ESC_H,                  0,
125       0,                       -ESC_K,       0,                       -ESC_K,
126       0,                       0,       0,                       0,
127       0,                       0,       -ESC_N,                  0,
128       -ESC_P,                  -ESC_Q,       -ESC_P,                  -ESC_Q,
129       -ESC_R,                  -ESC_S,       -ESC_R,                  -ESC_S,
130       0,                       0,       0,                       0,
# Line 165  static const short int escapes[] = { Line 171  static const short int escapes[] = {
171  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
172  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
173  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
174  /*  D0 */   '}',     0, -ESC_K,       0,      0,     0,      0, -ESC_P,  /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
175  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
176  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
177  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
# Line 182  string is built from string macros so th Line 188  string is built from string macros so th
188  platforms. */  platforms. */
189    
190  typedef struct verbitem {  typedef struct verbitem {
191    int   len;    int   len;                 /* Length of verb name */
192    int   op;    int   op;                  /* Op when no arg, or -1 if arg mandatory */
193      int   op_arg;              /* Op when arg present, or -1 if not allowed */
194  } verbitem;  } verbitem;
195    
196  static const char verbnames[] =  static const char verbnames[] =
197      "\0"                       /* Empty name is a shorthand for MARK */
198      STRING_MARK0
199    STRING_ACCEPT0    STRING_ACCEPT0
200    STRING_COMMIT0    STRING_COMMIT0
201    STRING_F0    STRING_F0
# Line 196  static const char verbnames[] = Line 205  static const char verbnames[] =
205    STRING_THEN;    STRING_THEN;
206    
207  static const verbitem verbs[] = {  static const verbitem verbs[] = {
208    { 6, OP_ACCEPT },    { 0, -1,        OP_MARK },
209    { 6, OP_COMMIT },    { 4, -1,        OP_MARK },
210    { 1, OP_FAIL },    { 6, OP_ACCEPT, -1 },
211    { 4, OP_FAIL },    { 6, OP_COMMIT, -1 },
212    { 5, OP_PRUNE },    { 1, OP_FAIL,   -1 },
213    { 4, OP_SKIP  },    { 4, OP_FAIL,   -1 },
214    { 4, OP_THEN  }    { 5, OP_PRUNE,  OP_PRUNE_ARG },
215      { 4, OP_SKIP,   OP_SKIP_ARG  },
216      { 4, OP_THEN,   OP_THEN_ARG  }
217  };  };
218    
219  static const int verbcount = sizeof(verbs)/sizeof(verbitem);  static const int verbcount = sizeof(verbs)/sizeof(verbitem);
# Line 250  static const int posix_class_maps[] = { Line 261  static const int posix_class_maps[] = {
261    cbit_xdigit,-1,          0              /* xdigit */    cbit_xdigit,-1,          0              /* xdigit */
262  };  };
263    
264    /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265    substitutes must be in the order of the names, defined above, and there are
266    both positive and negative cases. NULL means no substitute. */
267    
268    #ifdef SUPPORT_UCP
269    static const uschar *substitutes[] = {
270      (uschar *)"\\P{Nd}",    /* \D */
271      (uschar *)"\\p{Nd}",    /* \d */
272      (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
273      (uschar *)"\\p{Xsp}",   /* \s */
274      (uschar *)"\\P{Xwd}",   /* \W */
275      (uschar *)"\\p{Xwd}"    /* \w */
276    };
277    
278    static const uschar *posix_substitutes[] = {
279      (uschar *)"\\p{L}",     /* alpha */
280      (uschar *)"\\p{Ll}",    /* lower */
281      (uschar *)"\\p{Lu}",    /* upper */
282      (uschar *)"\\p{Xan}",   /* alnum */
283      NULL,                   /* ascii */
284      (uschar *)"\\h",        /* blank */
285      NULL,                   /* cntrl */
286      (uschar *)"\\p{Nd}",    /* digit */
287      NULL,                   /* graph */
288      NULL,                   /* print */
289      NULL,                   /* punct */
290      (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
291      (uschar *)"\\p{Xwd}",   /* word */
292      NULL,                   /* xdigit */
293      /* Negated cases */
294      (uschar *)"\\P{L}",     /* ^alpha */
295      (uschar *)"\\P{Ll}",    /* ^lower */
296      (uschar *)"\\P{Lu}",    /* ^upper */
297      (uschar *)"\\P{Xan}",   /* ^alnum */
298      NULL,                   /* ^ascii */
299      (uschar *)"\\H",        /* ^blank */
300      NULL,                   /* ^cntrl */
301      (uschar *)"\\P{Nd}",    /* ^digit */
302      NULL,                   /* ^graph */
303      NULL,                   /* ^print */
304      NULL,                   /* ^punct */
305      (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
306      (uschar *)"\\P{Xwd}",   /* ^word */
307      NULL                    /* ^xdigit */
308    };
309    #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310    #endif
311    
312  #define STRING(a)  # a  #define STRING(a)  # a
313  #define XSTRING(s) STRING(s)  #define XSTRING(s) STRING(s)
# Line 262  the number of relocations needed when a Line 320  the number of relocations needed when a
320  it is now one long string. We cannot use a table of offsets, because the  it is now one long string. We cannot use a table of offsets, because the
321  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we  lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322  simply count through to the one we want - this isn't a performance issue  simply count through to the one we want - this isn't a performance issue
323  because these strings are used only when there is a compilation error. */  because these strings are used only when there is a compilation error.
324    
325    Each substring ends with \0 to insert a null character. This includes the final
326    substring, so that the whole string ends with \0\0, which can be detected when
327    counting through. */
328    
329  static const char error_texts[] =  static const char error_texts[] =
330    "no error\0"    "no error\0"
# Line 309  static const char error_texts[] = Line 371  static const char error_texts[] =
371    /* 35 */    /* 35 */
372    "invalid condition (?(0)\0"    "invalid condition (?(0)\0"
373    "\\C not allowed in lookbehind assertion\0"    "\\C not allowed in lookbehind assertion\0"
374    "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"    "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375    "number after (?C is > 255\0"    "number after (?C is > 255\0"
376    "closing ) for (?C expected\0"    "closing ) for (?C expected\0"
377    /* 40 */    /* 40 */
# Line 331  static const char error_texts[] = Line 393  static const char error_texts[] =
393    "internal error: previously-checked referenced subpattern not found\0"    "internal error: previously-checked referenced subpattern not found\0"
394    "DEFINE group contains more than one branch\0"    "DEFINE group contains more than one branch\0"
395    /* 55 */    /* 55 */
396    "repeating a DEFINE group is not allowed\0"    "repeating a DEFINE group is not allowed\0"  /** DEAD **/
397    "inconsistent NEWLINE options\0"    "inconsistent NEWLINE options\0"
398    "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"    "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399    "a numbered reference must not be zero\0"    "a numbered reference must not be zero\0"
400    "(*VERB) with an argument is not supported\0"    "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401    /* 60 */    /* 60 */
402    "(*VERB) not recognized\0"    "(*VERB) not recognized\0"
403    "number is too big\0"    "number is too big\0"
404    "subpattern name expected\0"    "subpattern name expected\0"
405    "digit expected after (?+\0"    "digit expected after (?+\0"
406    "] is an invalid data character in JavaScript compatibility mode";    "] is an invalid data character in JavaScript compatibility mode\0"
407      /* 65 */
408      "different names for subpatterns of the same number are not allowed\0"
409      "(*MARK) must have an argument\0"
410      "this version of PCRE is not compiled with PCRE_UCP support\0"
411      "\\c must be followed by an ASCII character\0"
412      "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413      ;
414    
415  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
416  patterns. Note that the tables in chartables are dependent on the locale, and  patterns. Note that the tables in chartables are dependent on the locale, and
# Line 478  static const unsigned char ebcdic_charta Line 546  static const unsigned char ebcdic_charta
546  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
547    
548  static BOOL  static BOOL
549    compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
550      int *, int *, branch_chain *, compile_data *, int *);      int *, branch_chain *, compile_data *, int *);
551    
552    
553    
# Line 500  static const char * Line 568  static const char *
568  find_error_text(int n)  find_error_text(int n)
569  {  {
570  const char *s = error_texts;  const char *s = error_texts;
571  for (; n > 0; n--) while (*s++ != 0) {};  for (; n > 0; n--)
572      {
573      while (*s++ != 0) {};
574      if (*s == 0) return "Error text not found (please report)";
575      }
576  return s;  return s;
577  }  }
578    
# Line 571  else Line 643  else
643    
644      case CHAR_l:      case CHAR_l:
645      case CHAR_L:      case CHAR_L:
     case CHAR_N:  
646      case CHAR_u:      case CHAR_u:
647      case CHAR_U:      case CHAR_U:
648      *errorcodeptr = ERR37;      *errorcodeptr = ERR37;
# Line 772  else Line 843  else
843      break;      break;
844    
845      /* 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.
846      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
847        coding is ASCII-specific, but then the whole concept of \cx is
848      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */      ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
849    
850      case CHAR_c:      case CHAR_c:
# Line 782  else Line 854  else
854        *errorcodeptr = ERR2;        *errorcodeptr = ERR2;
855        break;        break;
856        }        }
857    #ifndef EBCDIC    /* ASCII/UTF-8 coding */
858  #ifndef EBCDIC  /* ASCII/UTF-8 coding */      if (c > 127)  /* Excludes all non-ASCII in either mode */
859          {
860          *errorcodeptr = ERR68;
861          break;
862          }
863      if (c >= CHAR_a && c <= CHAR_z) c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
864      c ^= 0x40;      c ^= 0x40;
865  #else           /* EBCDIC coding */  #else             /* EBCDIC coding */
866      if (c >= CHAR_a && c <= CHAR_z) c += 64;      if (c >= CHAR_a && c <= CHAR_z) c += 64;
867      c ^= 0xC0;      c ^= 0xC0;
868  #endif  #endif
# Line 809  else Line 885  else
885      }      }
886    }    }
887    
888    /* Perl supports \N{name} for character names, as well as plain \N for "not
889    newline". PCRE does not support \N{name}. */
890    
891    if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
892      *errorcodeptr = ERR37;
893    
894    /* If PCRE_UCP is set, we change the values for \d etc. */
895    
896    if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
897      c -= (ESC_DU - ESC_D);
898    
899    /* Set the pointer to the final character before returning. */
900    
901  *ptrptr = ptr;  *ptrptr = ptr;
902  return c;  return c;
903  }  }
# Line 1017  top-level call starts at the beginning o Line 1106  top-level call starts at the beginning o
1106  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
1107  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
1108  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
1109  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
1110  encountered, the name will be terminated by '>' because that is checked in the  track of subpatterns that reset the capturing group numbers - the (?| feature.
1111  first pass. Recursion is used to keep track of subpatterns that reset the  
1112  capturing group numbers - the (?| feature.  This function was originally called only from the second pass, in which we know
1113    that if (?< or (?' or (?P< is encountered, the name will be correctly
1114    terminated because that is checked in the first pass. There is now one call to
1115    this function in the first pass, to check for a recursive back reference by
1116    name (so that we can make the whole group atomic). In this case, we need check
1117    only up to the current position in the pattern, and that is still OK because
1118    and previous occurrences will have been checked. To make this work, the test
1119    for "end of pattern" is a check against cd->end_pattern in the main loop,
1120    instead of looking for a binary zero. This means that the special first-pass
1121    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1122    processing items within the loop are OK, because afterwards the main loop will
1123    terminate.)
1124    
1125  Arguments:  Arguments:
1126    ptrptr       address of the current character pointer (updated)    ptrptr       address of the current character pointer (updated)
# Line 1028  Arguments: Line 1128  Arguments:
1128    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1129    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1130    xmode        TRUE if we are in /x mode    xmode        TRUE if we are in /x mode
1131      utf8         TRUE if we are in UTF-8 mode
1132    count        pointer to the current capturing subpattern number (updated)    count        pointer to the current capturing subpattern number (updated)
1133    
1134  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 1136  Returns: the number of the named s
1136    
1137  static int  static int
1138  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,
1139    BOOL xmode, int *count)    BOOL xmode, BOOL utf8, int *count)
1140  {  {
1141  uschar *ptr = *ptrptr;  uschar *ptr = *ptrptr;
1142  int start_count = *count;  int start_count = *count;
# Line 1047  dealing with. The very first call may no Line 1148  dealing with. The very first call may no
1148    
1149  if (ptr[0] == CHAR_LEFT_PARENTHESIS)  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1150    {    {
1151    if (ptr[1] == CHAR_QUESTION_MARK &&    /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1152        ptr[2] == CHAR_VERTICAL_LINE)  
1153      if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1154    
1155      /* Handle a normal, unnamed capturing parenthesis. */
1156    
1157      else if (ptr[1] != CHAR_QUESTION_MARK)
1158        {
1159        *count += 1;
1160        if (name == NULL && *count == lorn) return *count;
1161        ptr++;
1162        }
1163    
1164      /* All cases now have (? at the start. Remember when we are in a group
1165      where the parenthesis numbers are duplicated. */
1166    
1167      else if (ptr[2] == CHAR_VERTICAL_LINE)
1168      {      {
1169      ptr += 3;      ptr += 3;
1170      dup_parens = TRUE;      dup_parens = TRUE;
1171      }      }
1172    
1173    /* Handle a normal, unnamed capturing parenthesis */    /* Handle comments; all characters are allowed until a ket is reached. */
1174    
1175    else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)    else if (ptr[2] == CHAR_NUMBER_SIGN)
1176      {      {
1177      *count += 1;      for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1178      if (name == NULL && *count == lorn) return *count;      goto FAIL_EXIT;
     ptr++;  
1179      }      }
1180    
1181    /* 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
1182    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
1183    condition (there can't be any nested parens. */    condition (there can't be any nested parens). */
1184    
1185    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1186      {      {
# Line 1077  if (ptr[0] == CHAR_LEFT_PARENTHESIS) Line 1192  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1192        }        }
1193      }      }
1194    
1195    /* We have either (? or (* and not a condition */    /* Start with (? but not a condition. */
1196    
1197    else    else
1198      {      {
# Line 1100  if (ptr[0] == CHAR_LEFT_PARENTHESIS) Line 1215  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1215        if (name != NULL && lorn == ptr - thisname &&        if (name != NULL && lorn == ptr - thisname &&
1216            strncmp((const char *)name, (const char *)thisname, lorn) == 0)            strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1217          return *count;          return *count;
1218        term++;        term++;
1219        }        }
1220      }      }
1221    }    }
1222    
1223  /* Past any initial parenthesis handling, scan for parentheses or vertical  /* Past any initial parenthesis handling, scan for parentheses or vertical
1224  bars. */  bars. Stop if we get to cd->end_pattern. Note that this is important for the
1225    first-pass call when this value is temporarily adjusted to stop at the current
1226    position. So DO NOT change this to a test for binary zero. */
1227    
1228  for (; *ptr != 0; ptr++)  for (; ptr < cd->end_pattern; ptr++)
1229    {    {
1230    /* Skip over backslashed characters and also entire \Q...\E */    /* Skip over backslashed characters and also entire \Q...\E */
1231    
# Line 1146  for (; *ptr != 0; ptr++) Line 1263  for (; *ptr != 0; ptr++)
1263            break;            break;
1264          }          }
1265        else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)        else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1266          {          {
1267          negate_class = TRUE;          negate_class = TRUE;
1268          ptr++;          ptr++;
1269          }          }
1270        else break;        else break;
1271        }        }
1272    
# Line 1182  for (; *ptr != 0; ptr++) Line 1299  for (; *ptr != 0; ptr++)
1299    
1300    if (xmode && *ptr == CHAR_NUMBER_SIGN)    if (xmode && *ptr == CHAR_NUMBER_SIGN)
1301      {      {
1302      while (*(++ptr) != 0 && *ptr != CHAR_NL) {};      ptr++;
1303        while (*ptr != 0)
1304          {
1305          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1306          ptr++;
1307    #ifdef SUPPORT_UTF8
1308          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1309    #endif
1310          }
1311      if (*ptr == 0) goto FAIL_EXIT;      if (*ptr == 0) goto FAIL_EXIT;
1312      continue;      continue;
1313      }      }
# Line 1191  for (; *ptr != 0; ptr++) Line 1316  for (; *ptr != 0; ptr++)
1316    
1317    if (*ptr == CHAR_LEFT_PARENTHESIS)    if (*ptr == CHAR_LEFT_PARENTHESIS)
1318      {      {
1319      int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);      int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1320      if (rc > 0) return rc;      if (rc > 0) return rc;
1321      if (*ptr == 0) goto FAIL_EXIT;      if (*ptr == 0) goto FAIL_EXIT;
1322      }      }
# Line 1199  for (; *ptr != 0; ptr++) Line 1324  for (; *ptr != 0; ptr++)
1324    else if (*ptr == CHAR_RIGHT_PARENTHESIS)    else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1325      {      {
1326      if (dup_parens && *count < hwm_count) *count = hwm_count;      if (dup_parens && *count < hwm_count) *count = hwm_count;
1327      *ptrptr = ptr;      goto FAIL_EXIT;
     return -1;  
1328      }      }
1329    
1330    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)    else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
# Line 1238  Arguments: Line 1362  Arguments:
1362    name         name to seek, or NULL if seeking a numbered subpattern    name         name to seek, or NULL if seeking a numbered subpattern
1363    lorn         name length, or subpattern number if name is NULL    lorn         name length, or subpattern number if name is NULL
1364    xmode        TRUE if we are in /x mode    xmode        TRUE if we are in /x mode
1365      utf8         TRUE if we are in UTF-8 mode
1366    
1367  Returns:       the number of the found subpattern, or -1 if not found  Returns:       the number of the found subpattern, or -1 if not found
1368  */  */
1369    
1370  static int  static int
1371  find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)  find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1372      BOOL utf8)
1373  {  {
1374  uschar *ptr = (uschar *)cd->start_pattern;  uschar *ptr = (uschar *)cd->start_pattern;
1375  int count = 0;  int count = 0;
# Line 1256  matching closing parens. That is why we Line 1382  matching closing parens. That is why we
1382    
1383  for (;;)  for (;;)
1384    {    {
1385    rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);    rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1386    if (rc > 0 || *ptr++ == 0) break;    if (rc > 0 || *ptr++ == 0) break;
1387    }    }
1388    
# Line 1272  return rc; Line 1398  return rc;
1398    
1399  /* This is called by several functions that scan a compiled expression looking  /* This is called by several functions that scan a compiled expression looking
1400  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
1401  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
1402  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
1403  assertions, and also the \b assertion; for others it does not.  does not.
1404    
1405  Arguments:  Arguments:
1406    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  
1407    skipassert   TRUE if certain assertions are to be skipped    skipassert   TRUE if certain assertions are to be skipped
1408    
1409  Returns:       pointer to the first significant opcode  Returns:       pointer to the first significant opcode
1410  */  */
1411    
1412  static const uschar*  static const uschar*
1413  first_significant_code(const uschar *code, int *options, int optbit,  first_significant_code(const uschar *code, BOOL skipassert)
   BOOL skipassert)  
1414  {  {
1415  for (;;)  for (;;)
1416    {    {
1417    switch ((int)*code)    switch ((int)*code)
1418      {      {
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
   
1419      case OP_ASSERT_NOT:      case OP_ASSERT_NOT:
1420      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1421      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
# Line 1315  for (;;) Line 1431  for (;;)
1431    
1432      case OP_CALLOUT:      case OP_CALLOUT:
1433      case OP_CREF:      case OP_CREF:
1434        case OP_NCREF:
1435      case OP_RREF:      case OP_RREF:
1436        case OP_NRREF:
1437      case OP_DEF:      case OP_DEF:
1438      code += _pcre_OP_lengths[*code];      code += _pcre_OP_lengths[*code];
1439      break;      break;
# Line 1336  for (;;) Line 1454  for (;;)
1454    
1455  /* 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,
1456  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.
1457  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
1458  temporarily terminated with OP_END when this function is called.  temporarily terminated with OP_END when this function is called.
1459    
1460  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
1461  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.
1462  However, we cannot do this when the assertion contains subroutine calls,  However, we cannot do this when the assertion contains subroutine calls,
1463  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
1464  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.
1465    
1466  Arguments:  Arguments:
1467    code     points to the start of the pattern (the bracket)    code     points to the start of the pattern (the bracket)
1468    options  the compiling options    utf8     TRUE in UTF-8 mode
1469    atend    TRUE if called when the pattern is complete    atend    TRUE if called when the pattern is complete
1470    cd       the "compile data" structure    cd       the "compile data" structure
1471    
1472  Returns:   the fixed length,  Returns:   the fixed length,
1473               or -1 if there is no fixed length,               or -1 if there is no fixed length,
1474               or -2 if \C was encountered               or -2 if \C was encountered
1475               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
1476  */  */
1477    
1478  static int  static int
1479  find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1480  {  {
1481  int length = -1;  int length = -1;
1482    
# Line 1375  for (;;) Line 1493  for (;;)
1493    register int op = *cc;    register int op = *cc;
1494    switch (op)    switch (op)
1495      {      {
1496        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1497        OP_BRA (normal non-capturing bracket) because the other variants of these
1498        opcodes are all concerned with unlimited repeated groups, which of course
1499        are not of fixed length. They will cause a -1 response from the default
1500        case of this switch. */
1501    
1502      case OP_CBRA:      case OP_CBRA:
1503      case OP_BRA:      case OP_BRA:
1504      case OP_ONCE:      case OP_ONCE:
1505      case OP_COND:      case OP_COND:
1506      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1507      if (d < 0) return d;      if (d < 0) return d;
1508      branchlength += d;      branchlength += d;
1509      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
# Line 1388  for (;;) Line 1512  for (;;)
1512    
1513      /* 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
1514      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
1515      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.
1516        Note that we must not include the OP_KETRxxx opcodes here, because they
1517        all imply an unlimited repeat. */
1518    
1519      case OP_ALT:      case OP_ALT:
1520      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1521      case OP_END:      case OP_END:
1522      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1523        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
# Line 1401  for (;;) Line 1525  for (;;)
1525      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1526      branchlength = 0;      branchlength = 0;
1527      break;      break;
1528    
1529      /* A true recursion implies not fixed length, but a subroutine call may      /* A true recursion implies not fixed length, but a subroutine call may
1530      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
1531      it until the end of the pattern, so return -3. */      it until the end of the pattern, so return -3. */
1532    
1533      case OP_RECURSE:      case OP_RECURSE:
1534      if (!atend) return -3;      if (!atend) return -3;
1535      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */      cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1536      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */      do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1537      if (cc > cs && cc < ce) return -1;                /* Recursion */      if (cc > cs && cc < ce) return -1;                /* Recursion */
1538      d = find_fixedlength(cs + 2, options, atend, cd);      d = find_fixedlength(cs + 2, utf8, atend, cd);
1539      if (d < 0) return d;      if (d < 0) return d;
1540      branchlength += d;      branchlength += d;
1541      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1542      break;      break;
1543    
1544      /* Skip over assertive subpatterns */      /* Skip over assertive subpatterns */
1545    
# Line 1430  for (;;) Line 1554  for (;;)
1554    
1555      case OP_REVERSE:      case OP_REVERSE:
1556      case OP_CREF:      case OP_CREF:
1557        case OP_NCREF:
1558      case OP_RREF:      case OP_RREF:
1559        case OP_NRREF:
1560      case OP_DEF:      case OP_DEF:
     case OP_OPT:  
1561      case OP_CALLOUT:      case OP_CALLOUT:
1562      case OP_SOD:      case OP_SOD:
1563      case OP_SOM:      case OP_SOM:
1564        case OP_SET_SOM:
1565      case OP_EOD:      case OP_EOD:
1566      case OP_EODN:      case OP_EODN:
1567      case OP_CIRC:      case OP_CIRC:
1568        case OP_CIRCM:
1569      case OP_DOLL:      case OP_DOLL:
1570        case OP_DOLLM:
1571      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1572      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1573      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
# Line 1448  for (;;) Line 1576  for (;;)
1576      /* Handle literal characters */      /* Handle literal characters */
1577    
1578      case OP_CHAR:      case OP_CHAR:
1579      case OP_CHARNC:      case OP_CHARI:
1580      case OP_NOT:      case OP_NOT:
1581        case OP_NOTI:
1582      branchlength++;      branchlength++;
1583      cc += 2;      cc += 2;
1584  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1585      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];  
1586  #endif  #endif
1587      break;      break;
1588    
# Line 1465  for (;;) Line 1593  for (;;)
1593      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1594      cc += 4;      cc += 4;
1595  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1596      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];  
1597  #endif  #endif
1598      break;      break;
1599    
# Line 1550  for (;;) Line 1677  for (;;)
1677    
1678  /* This little function scans through a compiled pattern until it finds a  /* This little function scans through a compiled pattern until it finds a
1679  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
1680  instance of OP_REVERSE for a lookbehind. The function is global in the C sense  instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1681  so that it can be called from pcre_study() when finding the minimum matching  so that it can be called from pcre_study() when finding the minimum matching
1682  length.  length.
1683    
1684  Arguments:  Arguments:
# Line 1568  _pcre_find_bracket(const uschar *code, B Line 1695  _pcre_find_bracket(const uschar *code, B
1695  for (;;)  for (;;)
1696    {    {
1697    register int c = *code;    register int c = *code;
1698    
1699    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1700    
1701    /* 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 1575  for (;;) Line 1703  for (;;)
1703    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. */
1704    
1705    if (c == OP_XCLASS) code += GET(code, 1);    if (c == OP_XCLASS) code += GET(code, 1);
1706    
1707    /* Handle recursion */    /* Handle recursion */
1708    
1709    else if (c == OP_REVERSE)    else if (c == OP_REVERSE)
1710      {      {
1711      if (number < 0) return (uschar *)code;      if (number < 0) return (uschar *)code;
1712      code += _pcre_OP_lengths[c];      code += _pcre_OP_lengths[c];
1713      }      }
1714    
1715    /* Handle capturing bracket */    /* Handle capturing bracket */
1716    
1717    else if (c == OP_CBRA)    else if (c == OP_CBRA || c == OP_SCBRA ||
1718               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1719      {      {
1720      int n = GET2(code, 1+LINK_SIZE);      int n = GET2(code, 1+LINK_SIZE);
1721      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
# Line 1595  for (;;) Line 1724  for (;;)
1724    
1725    /* 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
1726    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
1727    two bytes of parameters. */    two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1728      must add in its length. */
1729    
1730    else    else
1731      {      {
# Line 1619  for (;;) Line 1749  for (;;)
1749        case OP_TYPEPOSUPTO:        case OP_TYPEPOSUPTO:
1750        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1751        break;        break;
1752    
1753          case OP_MARK:
1754          case OP_PRUNE_ARG:
1755          case OP_SKIP_ARG:
1756          code += code[1];
1757          break;
1758    
1759          case OP_THEN_ARG:
1760          code += code[1+LINK_SIZE];
1761          break;
1762        }        }
1763    
1764      /* Add in the fixed length from the table */      /* Add in the fixed length from the table */
# Line 1633  for (;;) Line 1773  for (;;)
1773      if (utf8) switch(c)      if (utf8) switch(c)
1774        {        {
1775        case OP_CHAR:        case OP_CHAR:
1776        case OP_CHARNC:        case OP_CHARI:
1777        case OP_EXACT:        case OP_EXACT:
1778          case OP_EXACTI:
1779        case OP_UPTO:        case OP_UPTO:
1780          case OP_UPTOI:
1781        case OP_MINUPTO:        case OP_MINUPTO:
1782          case OP_MINUPTOI:
1783        case OP_POSUPTO:        case OP_POSUPTO:
1784          case OP_POSUPTOI:
1785        case OP_STAR:        case OP_STAR:
1786          case OP_STARI:
1787        case OP_MINSTAR:        case OP_MINSTAR:
1788          case OP_MINSTARI:
1789        case OP_POSSTAR:        case OP_POSSTAR:
1790          case OP_POSSTARI:
1791        case OP_PLUS:        case OP_PLUS:
1792          case OP_PLUSI:
1793        case OP_MINPLUS:        case OP_MINPLUS:
1794          case OP_MINPLUSI:
1795        case OP_POSPLUS:        case OP_POSPLUS:
1796          case OP_POSPLUSI:
1797        case OP_QUERY:        case OP_QUERY:
1798          case OP_QUERYI:
1799        case OP_MINQUERY:        case OP_MINQUERY:
1800          case OP_MINQUERYI:
1801        case OP_POSQUERY:        case OP_POSQUERY:
1802          case OP_POSQUERYI:
1803        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1804        break;        break;
1805        }        }
# Line 1690  for (;;) Line 1843  for (;;)
1843    
1844    /* 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
1845    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
1846    two bytes of parameters. */    two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1847      must add in its length. */
1848    
1849    else    else
1850      {      {
# Line 1714  for (;;) Line 1868  for (;;)
1868        case OP_TYPEEXACT:        case OP_TYPEEXACT:
1869        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1870        break;        break;
1871    
1872          case OP_MARK:
1873          case OP_PRUNE_ARG:
1874          case OP_SKIP_ARG:
1875          code += code[1];
1876          break;
1877    
1878          case OP_THEN_ARG:
1879          code += code[1+LINK_SIZE];
1880          break;
1881        }        }
1882    
1883      /* Add in the fixed length from the table */      /* Add in the fixed length from the table */
# Line 1728  for (;;) Line 1892  for (;;)
1892      if (utf8) switch(c)      if (utf8) switch(c)
1893        {        {
1894        case OP_CHAR:        case OP_CHAR:
1895        case OP_CHARNC:        case OP_CHARI:
1896        case OP_EXACT:        case OP_EXACT:
1897          case OP_EXACTI:
1898        case OP_UPTO:        case OP_UPTO:
1899          case OP_UPTOI:
1900        case OP_MINUPTO:        case OP_MINUPTO:
1901          case OP_MINUPTOI:
1902        case OP_POSUPTO:        case OP_POSUPTO:
1903          case OP_POSUPTOI:
1904        case OP_STAR:        case OP_STAR:
1905          case OP_STARI:
1906        case OP_MINSTAR:        case OP_MINSTAR:
1907          case OP_MINSTARI:
1908        case OP_POSSTAR:        case OP_POSSTAR:
1909          case OP_POSSTARI:
1910        case OP_PLUS:        case OP_PLUS:
1911          case OP_PLUSI:
1912        case OP_MINPLUS:        case OP_MINPLUS:
1913          case OP_MINPLUSI:
1914        case OP_POSPLUS:        case OP_POSPLUS:
1915          case OP_POSPLUSI:
1916        case OP_QUERY:        case OP_QUERY:
1917          case OP_QUERYI:
1918        case OP_MINQUERY:        case OP_MINQUERY:
1919          case OP_MINQUERYI:
1920        case OP_POSQUERY:        case OP_POSQUERY:
1921          case OP_POSQUERYI:
1922        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1923        break;        break;
1924        }        }
# Line 1770  Arguments: Line 1947  Arguments:
1947    code        points to start of search    code        points to start of search
1948    endcode     points to where to stop    endcode     points to where to stop
1949    utf8        TRUE if in UTF8 mode    utf8        TRUE if in UTF8 mode
1950      cd          contains pointers to tables etc.
1951    
1952  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
1953  */  */
1954    
1955  static BOOL  static BOOL
1956  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1957      compile_data *cd)
1958  {  {
1959  register int c;  register int c;
1960  for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1961       code < endcode;       code < endcode;
1962       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1963    {    {
1964    const uschar *ccode;    const uschar *ccode;
1965    
# Line 1796  for (code = first_significant_code(code Line 1975  for (code = first_significant_code(code
1975      continue;      continue;
1976      }      }
1977    
1978      /* For a recursion/subroutine call, if its end has been reached, which
1979      implies a backward reference subroutine call, we can scan it. If it's a
1980      forward reference subroutine call, we can't. To detect forward reference
1981      we have to scan up the list that is kept in the workspace. This function is
1982      called only when doing the real compile, not during the pre-compile that
1983      measures the size of the compiled pattern. */
1984    
1985      if (c == OP_RECURSE)
1986        {
1987        const uschar *scode;
1988        BOOL empty_branch;
1989    
1990        /* Test for forward reference */
1991    
1992        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1993          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1994    
1995        /* Not a forward reference, test for completed backward reference */
1996    
1997        empty_branch = FALSE;
1998        scode = cd->start_code + GET(code, 1);
1999        if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2000    
2001        /* Completed backwards reference */
2002    
2003        do
2004          {
2005          if (could_be_empty_branch(scode, endcode, utf8, cd))
2006            {
2007            empty_branch = TRUE;
2008            break;
2009            }
2010          scode += GET(scode, 1);
2011          }
2012        while (*scode == OP_ALT);
2013    
2014        if (!empty_branch) return FALSE;  /* All branches are non-empty */
2015        continue;
2016        }
2017    
2018    /* Groups with zero repeats can of course be empty; skip them. */    /* Groups with zero repeats can of course be empty; skip them. */
2019    
2020    if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)    if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2021          c == OP_BRAPOSZERO)
2022      {      {
2023      code += _pcre_OP_lengths[c];      code += _pcre_OP_lengths[c];
2024      do code += GET(code, 1); while (*code == OP_ALT);      do code += GET(code, 1); while (*code == OP_ALT);
# Line 1806  for (code = first_significant_code(code Line 2026  for (code = first_significant_code(code
2026      continue;      continue;
2027      }      }
2028    
2029      /* A nested group that is already marked as "could be empty" can just be
2030      skipped. */
2031    
2032      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2033          c == OP_SCBRA || c == OP_SCBRAPOS)
2034        {
2035        do code += GET(code, 1); while (*code == OP_ALT);
2036        c = *code;
2037        continue;
2038        }
2039    
2040    /* For other groups, scan the branches. */    /* For other groups, scan the branches. */
2041    
2042    if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)    if (c == OP_BRA  || c == OP_BRAPOS ||
2043          c == OP_CBRA || c == OP_CBRAPOS ||
2044          c == OP_ONCE || c == OP_COND)
2045      {      {
2046      BOOL empty_branch;      BOOL empty_branch;
2047      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
# Line 1824  for (code = first_significant_code(code Line 2057  for (code = first_significant_code(code
2057        empty_branch = FALSE;        empty_branch = FALSE;
2058        do        do
2059          {          {
2060          if (!empty_branch && could_be_empty_branch(code, endcode, utf8))          if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2061            empty_branch = TRUE;            empty_branch = TRUE;
2062          code += GET(code, 1);          code += GET(code, 1);
2063          }          }
# Line 1895  for (code = first_significant_code(code Line 2128  for (code = first_significant_code(code
2128      case OP_ALLANY:      case OP_ALLANY:
2129      case OP_ANYBYTE:      case OP_ANYBYTE:
2130      case OP_CHAR:      case OP_CHAR:
2131      case OP_CHARNC:      case OP_CHARI:
2132      case OP_NOT:      case OP_NOT:
2133        case OP_NOTI:
2134      case OP_PLUS:      case OP_PLUS:
2135      case OP_MINPLUS:      case OP_MINPLUS:
2136      case OP_POSPLUS:      case OP_POSPLUS:
# Line 1936  for (code = first_significant_code(code Line 2170  for (code = first_significant_code(code
2170      case OP_KET:      case OP_KET:
2171      case OP_KETRMAX:      case OP_KETRMAX:
2172      case OP_KETRMIN:      case OP_KETRMIN:
2173        case OP_KETRPOS:
2174      case OP_ALT:      case OP_ALT:
2175      return TRUE;      return TRUE;
2176    
# Line 1944  for (code = first_significant_code(code Line 2179  for (code = first_significant_code(code
2179    
2180  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2181      case OP_STAR:      case OP_STAR:
2182        case OP_STARI:
2183      case OP_MINSTAR:      case OP_MINSTAR:
2184        case OP_MINSTARI:
2185      case OP_POSSTAR:      case OP_POSSTAR:
2186        case OP_POSSTARI:
2187      case OP_QUERY:      case OP_QUERY:
2188        case OP_QUERYI:
2189      case OP_MINQUERY:      case OP_MINQUERY:
2190        case OP_MINQUERYI:
2191      case OP_POSQUERY:      case OP_POSQUERY:
2192        case OP_POSQUERYI:
2193      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];      if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2194      break;      break;
2195    
2196      case OP_UPTO:      case OP_UPTO:
2197        case OP_UPTOI:
2198      case OP_MINUPTO:      case OP_MINUPTO:
2199        case OP_MINUPTOI:
2200      case OP_POSUPTO:      case OP_POSUPTO:
2201        case OP_POSUPTOI:
2202      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];      if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2203      break;      break;
2204  #endif  #endif
2205    
2206        /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2207        string. */
2208    
2209        case OP_MARK:
2210        case OP_PRUNE_ARG:
2211        case OP_SKIP_ARG:
2212        code += code[1];
2213        break;
2214    
2215        case OP_THEN_ARG:
2216        code += code[1+LINK_SIZE];
2217        break;
2218    
2219        /* None of the remaining opcodes are required to match a character. */
2220    
2221        default:
2222        break;
2223      }      }
2224    }    }
2225    
# Line 1974  return TRUE; Line 2236  return TRUE;
2236  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
2237  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,
2238  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.
2239    This function is called only during the real compile, not during the
2240    pre-compile.
2241    
2242  Arguments:  Arguments:
2243    code        points to start of the recursion    code        points to start of the recursion
2244    endcode     points to where to stop (current RECURSE item)    endcode     points to where to stop (current RECURSE item)
2245    bcptr       points to the chain of current (unclosed) branch starts    bcptr       points to the chain of current (unclosed) branch starts
2246    utf8        TRUE if in UTF-8 mode    utf8        TRUE if in UTF-8 mode
2247      cd          pointers to tables etc
2248    
2249  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2250  */  */
2251    
2252  static BOOL  static BOOL
2253  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2254    BOOL utf8)    BOOL utf8, compile_data *cd)
2255  {  {
2256  while (bcptr != NULL && bcptr->current >= code)  while (bcptr != NULL && bcptr->current_branch >= code)
2257    {    {
2258    if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;    if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2259        return FALSE;
2260    bcptr = bcptr->outer;    bcptr = bcptr->outer;
2261    }    }
2262  return TRUE;  return TRUE;
# Line 2172  auto_callout(uschar *code, const uschar Line 2438  auto_callout(uschar *code, const uschar
2438  {  {
2439  *code++ = OP_CALLOUT;  *code++ = OP_CALLOUT;
2440  *code++ = 255;  *code++ = 255;
2441  PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
2442  PUT(code, LINK_SIZE, 0);                /* Default length */  PUT(code, LINK_SIZE, 0);                       /* Default length */
2443  return code + 2*LINK_SIZE;  return code + 2*LINK_SIZE;
2444  }  }
2445    
# Line 2198  Returns: nothing Line 2464  Returns: nothing
2464  static void  static void
2465  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2466  {  {
2467  int length = ptr - cd->start_pattern - GET(previous_callout, 2);  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2468  PUT(previous_callout, 2 + LINK_SIZE, length);  PUT(previous_callout, 2 + LINK_SIZE, length);
2469  }  }
2470    
# Line 2248  for (++c; c <= d; c++) Line 2514  for (++c; c <= d; c++)
2514    
2515  return TRUE;  return TRUE;
2516  }  }
2517    
2518    
2519    
2520    /*************************************************
2521    *        Check a character and a property        *
2522    *************************************************/
2523    
2524    /* This function is called by check_auto_possessive() when a property item
2525    is adjacent to a fixed character.
2526    
2527    Arguments:
2528      c            the character
2529      ptype        the property type
2530      pdata        the data for the type
2531      negated      TRUE if it's a negated property (\P or \p{^)
2532    
2533    Returns:       TRUE if auto-possessifying is OK
2534    */
2535    
2536    static BOOL
2537    check_char_prop(int c, int ptype, int pdata, BOOL negated)
2538    {
2539    const ucd_record *prop = GET_UCD(c);
2540    switch(ptype)
2541      {
2542      case PT_LAMP:
2543      return (prop->chartype == ucp_Lu ||
2544              prop->chartype == ucp_Ll ||
2545              prop->chartype == ucp_Lt) == negated;
2546    
2547      case PT_GC:
2548      return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2549    
2550      case PT_PC:
2551      return (pdata == prop->chartype) == negated;
2552    
2553      case PT_SC:
2554      return (pdata == prop->script) == negated;
2555    
2556      /* These are specials */
2557    
2558      case PT_ALNUM:
2559      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2560              _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2561    
2562      case PT_SPACE:    /* Perl space */
2563      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2564              c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2565              == negated;
2566    
2567      case PT_PXSPACE:  /* POSIX space */
2568      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2569              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2570              c == CHAR_FF || c == CHAR_CR)
2571              == negated;
2572    
2573      case PT_WORD:
2574      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2575              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2576              c == CHAR_UNDERSCORE) == negated;
2577      }
2578    return FALSE;
2579    }
2580  #endif  /* SUPPORT_UCP */  #endif  /* SUPPORT_UCP */
2581    
2582    
# Line 2261  whether the next thing could possibly ma Line 2590  whether the next thing could possibly ma
2590  sense to automatically possessify the repeated item.  sense to automatically possessify the repeated item.
2591    
2592  Arguments:  Arguments:
2593    op_code       the repeated op code    previous      pointer to the repeated opcode
   this          data for this item, depends on the opcode  
2594    utf8          TRUE in UTF-8 mode    utf8          TRUE in UTF-8 mode
   utf8_char     used for utf8 character bytes, NULL if not relevant  
2595    ptr           next character in pattern    ptr           next character in pattern
2596    options       options bits    options       options bits
2597    cd            contains pointers to tables etc.    cd            contains pointers to tables etc.
# Line 2273  Returns: TRUE if possessifying is Line 2600  Returns: TRUE if possessifying is
2600  */  */
2601    
2602  static BOOL  static BOOL
2603  check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,  check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2604    const uschar *ptr, int options, compile_data *cd)    int options, compile_data *cd)
2605  {  {
2606  int next;  int c, next;
2607    int op_code = *previous++;
2608    
2609  /* Skip whitespace and comments in extended mode */  /* Skip whitespace and comments in extended mode */
2610    
# Line 2287  if ((options & PCRE_EXTENDED) != 0) Line 2615  if ((options & PCRE_EXTENDED) != 0)
2615      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2616      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2617        {        {
2618        while (*(++ptr) != 0)        ptr++;
2619          while (*ptr != 0)
2620            {
2621          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2622            ptr++;
2623    #ifdef SUPPORT_UTF8
2624            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2625    #endif
2626            }
2627        }        }
2628      else break;      else break;
2629      }      }
# Line 2324  if ((options & PCRE_EXTENDED) != 0) Line 2659  if ((options & PCRE_EXTENDED) != 0)
2659      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2660      if (*ptr == CHAR_NUMBER_SIGN)      if (*ptr == CHAR_NUMBER_SIGN)
2661        {        {
2662        while (*(++ptr) != 0)        ptr++;
2663          while (*ptr != 0)
2664            {
2665          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2666            ptr++;
2667    #ifdef SUPPORT_UTF8
2668            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2669    #endif
2670            }
2671        }        }
2672      else break;      else break;
2673      }      }
# Line 2337  if (*ptr == CHAR_ASTERISK || *ptr == CHA Line 2679  if (*ptr == CHAR_ASTERISK || *ptr == CHA
2679    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)
2680      return FALSE;      return FALSE;
2681    
2682  /* 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
2683  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. */  
2684    
2685  if (next >= 0) switch(op_code)  if (next >= 0) switch(op_code)
2686    {    {
2687    case OP_CHAR:    case OP_CHAR:
2688  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2689    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2690  #else  #else
2691    (void)(utf8_char);  /* Keep compiler happy by referencing function argument */    c = *previous;
2692  #endif  #endif
2693    return item != next;    return c != next;
2694    
2695    /* 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
2696    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
2697    high-valued characters. */    high-valued characters. */
2698    
2699    case OP_CHARNC:    case OP_CHARI:
2700  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2701    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2702    #else
2703      c = *previous;
2704  #endif  #endif
2705    if (item == next) return FALSE;    if (c == next) return FALSE;
2706  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2707    if (utf8)    if (utf8)
2708      {      {
# Line 2374  if (next >= 0) switch(op_code) Line 2713  if (next >= 0) switch(op_code)
2713  #else  #else
2714      othercase = NOTACHAR;      othercase = NOTACHAR;
2715  #endif  #endif
2716      return (unsigned int)item != othercase;      return (unsigned int)c != othercase;
2717      }      }
2718    else    else
2719  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2720    return (item != cd->fcc[next]);  /* Non-UTF-8 mode */    return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2721    
2722    /* 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
2723      opcodes are not used for multi-byte characters, because they are coded using
2724      an XCLASS instead. */
2725    
2726    case OP_NOT:    case OP_NOT:
2727    if (item == next) return TRUE;    return (c = *previous) == next;
2728    if ((options & PCRE_CASELESS) == 0) return FALSE;  
2729      case OP_NOTI:
2730      if ((c = *previous) == next) return TRUE;
2731  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2732    if (utf8)    if (utf8)
2733      {      {
# Line 2395  if (next >= 0) switch(op_code) Line 2738  if (next >= 0) switch(op_code)
2738  #else  #else
2739      othercase = NOTACHAR;      othercase = NOTACHAR;
2740  #endif  #endif
2741      return (unsigned int)item == othercase;      return (unsigned int)c == othercase;
2742      }      }
2743    else    else
2744  #endif  /* SUPPORT_UTF8 */  #endif  /* SUPPORT_UTF8 */
2745    return (item == cd->fcc[next]);  /* Non-UTF-8 mode */    return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
2746    
2747      /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2748      When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2749    
2750    case OP_DIGIT:    case OP_DIGIT:
2751    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
# Line 2442  if (next >= 0) switch(op_code) Line 2788  if (next >= 0) switch(op_code)
2788      case 0x202f:      case 0x202f:
2789      case 0x205f:      case 0x205f:
2790      case 0x3000:      case 0x3000:
2791      return op_code != OP_HSPACE;      return op_code == OP_NOT_HSPACE;
2792      default:      default:
2793      return op_code == OP_HSPACE;      return op_code != OP_NOT_HSPACE;
2794      }      }
2795    
2796      case OP_ANYNL:
2797    case OP_VSPACE:    case OP_VSPACE:
2798    case OP_NOT_VSPACE:    case OP_NOT_VSPACE:
2799    switch(next)    switch(next)
# Line 2458  if (next >= 0) switch(op_code) Line 2805  if (next >= 0) switch(op_code)
2805      case 0x85:      case 0x85:
2806      case 0x2028:      case 0x2028:
2807      case 0x2029:      case 0x2029:
2808      return op_code != OP_VSPACE;      return op_code == OP_NOT_VSPACE;
2809      default:      default:
2810      return op_code == OP_VSPACE;      return op_code != OP_NOT_VSPACE;
2811      }      }
2812    
2813    #ifdef SUPPORT_UCP
2814      case OP_PROP:
2815      return check_char_prop(next, previous[0], previous[1], FALSE);
2816    
2817      case OP_NOTPROP:
2818      return check_char_prop(next, previous[0], previous[1], TRUE);
2819    #endif
2820    
2821    default:    default:
2822    return FALSE;    return FALSE;
2823    }    }
2824    
2825    
2826  /* 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
2827    is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2828    generated only when PCRE_UCP is *not* set, that is, when only ASCII
2829    characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2830    replaced by OP_PROP codes when PCRE_UCP is set. */
2831    
2832  switch(op_code)  switch(op_code)
2833    {    {
2834    case OP_CHAR:    case OP_CHAR:
2835    case OP_CHARNC:    case OP_CHARI:
2836  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2837    if (utf8 && item > 127) { GETCHAR(item, utf8_char); }    GETCHARTEST(c, previous);
2838    #else
2839      c = *previous;
2840  #endif  #endif
2841    switch(-next)    switch(-next)
2842      {      {
2843      case ESC_d:      case ESC_d:
2844      return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;      return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2845    
2846      case ESC_D:      case ESC_D:
2847      return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2848    
2849      case ESC_s:      case ESC_s:
2850      return item > 127 || (cd->ctypes[item] & ctype_space) == 0;      return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2851    
2852      case ESC_S:      case ESC_S:
2853      return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2854    
2855      case ESC_w:      case ESC_w:
2856      return item > 127 || (cd->ctypes[item] & ctype_word) == 0;      return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2857    
2858      case ESC_W:      case ESC_W:
2859      return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;      return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2860    
2861      case ESC_h:      case ESC_h:
2862      case ESC_H:      case ESC_H:
2863      switch(item)      switch(c)
2864        {        {
2865        case 0x09:        case 0x09:
2866        case 0x20:        case 0x20:
# Line 2527  switch(op_code) Line 2888  switch(op_code)
2888    
2889      case ESC_v:      case ESC_v:
2890      case ESC_V:      case ESC_V:
2891      switch(item)      switch(c)
2892        {        {
2893        case 0x0a:        case 0x0a:
2894        case 0x0b:        case 0x0b:
# Line 2541  switch(op_code) Line 2902  switch(op_code)
2902        return -next == ESC_v;        return -next == ESC_v;
2903        }        }
2904    
2905        /* When PCRE_UCP is set, these values get generated for \d etc. Find
2906        their substitutions and process them. The result will always be either
2907        -ESC_p or -ESC_P. Then fall through to process those values. */
2908    
2909    #ifdef SUPPORT_UCP
2910        case ESC_du:
2911        case ESC_DU:
2912        case ESC_wu:
2913        case ESC_WU:
2914        case ESC_su:
2915        case ESC_SU:
2916          {
2917          int temperrorcode = 0;
2918          ptr = substitutes[-next - ESC_DU];
2919          next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2920          if (temperrorcode != 0) return FALSE;
2921          ptr++;    /* For compatibility */
2922          }
2923        /* Fall through */
2924    
2925        case ESC_p:
2926        case ESC_P:
2927          {
2928          int ptype, pdata, errorcodeptr;
2929          BOOL negated;
2930    
2931          ptr--;      /* Make ptr point at the p or P */
2932          ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2933          if (ptype < 0) return FALSE;
2934          ptr++;      /* Point past the final curly ket */
2935    
2936          /* If the property item is optional, we have to give up. (When generated
2937          from \d etc by PCRE_UCP, this test will have been applied much earlier,
2938          to the original \d etc. At this point, ptr will point to a zero byte. */
2939    
2940          if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2941            strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2942              return FALSE;
2943    
2944          /* Do the property check. */
2945    
2946          return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2947          }
2948    #endif
2949    
2950      default:      default:
2951      return FALSE;      return FALSE;
2952      }      }
2953    
2954      /* In principle, support for Unicode properties should be integrated here as
2955      well. It means re-organizing the above code so as to get hold of the property
2956      values before switching on the op-code. However, I wonder how many patterns
2957      combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2958      these op-codes are never generated.) */
2959    
2960    case OP_DIGIT:    case OP_DIGIT:
2961    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2962           next == -ESC_h || next == -ESC_v;           next == -ESC_h || next == -ESC_v || next == -ESC_R;
2963    
2964    case OP_NOT_DIGIT:    case OP_NOT_DIGIT:
2965    return next == -ESC_d;    return next == -ESC_d;
2966    
2967    case OP_WHITESPACE:    case OP_WHITESPACE:
2968    return next == -ESC_S || next == -ESC_d || next == -ESC_w;    return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2969    
2970    case OP_NOT_WHITESPACE:    case OP_NOT_WHITESPACE:
2971    return next == -ESC_s || next == -ESC_h || next == -ESC_v;    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2972    
2973    case OP_HSPACE:    case OP_HSPACE:
2974    return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2975             next == -ESC_w || next == -ESC_v || next == -ESC_R;
2976    
2977    case OP_NOT_HSPACE:    case OP_NOT_HSPACE:
2978    return next == -ESC_h;    return next == -ESC_h;
2979    
2980    /* Can't have \S in here because VT matches \S (Perl anomaly) */    /* Can't have \S in here because VT matches \S (Perl anomaly) */
2981      case OP_ANYNL:
2982    case OP_VSPACE:    case OP_VSPACE:
2983    return next == -ESC_V || next == -ESC_d || next == -ESC_w;    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2984    
2985    case OP_NOT_VSPACE:    case OP_NOT_VSPACE:
2986    return next == -ESC_v;    return next == -ESC_v || next == -ESC_R;
2987    
2988    case OP_WORDCHAR:    case OP_WORDCHAR:
2989    return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;    return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2990             next == -ESC_v || next == -ESC_R;
2991    
2992    case OP_NOT_WORDCHAR:    case OP_NOT_WORDCHAR:
2993    return next == -ESC_w || next == -ESC_d;    return next == -ESC_w || next == -ESC_d;
# Line 2624  int greedy_default, greedy_non_default; Line 3039  int greedy_default, greedy_non_default;
3039  int firstbyte, reqbyte;  int firstbyte, reqbyte;
3040  int zeroreqbyte, zerofirstbyte;  int zeroreqbyte, zerofirstbyte;
3041  int req_caseopt, reqvary, tempreqvary;  int req_caseopt, reqvary, tempreqvary;
3042  int options = *optionsptr;  int options = *optionsptr;               /* May change dynamically */
3043  int after_manual_callout = 0;  int after_manual_callout = 0;
3044  int length_prevgroup = 0;  int length_prevgroup = 0;
3045  register int c;  register int c;
# Line 2636  BOOL inescq = FALSE; Line 3051  BOOL inescq = FALSE;
3051  BOOL groupsetfirstbyte = FALSE;  BOOL groupsetfirstbyte = FALSE;
3052  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
3053  const uschar *tempptr;  const uschar *tempptr;
3054    const uschar *nestptr = NULL;
3055  uschar *previous = NULL;  uschar *previous = NULL;
3056  uschar *previous_callout = NULL;  uschar *previous_callout = NULL;
3057  uschar *save_hwm = NULL;  uschar *save_hwm = NULL;
3058  uschar classbits[32];  uschar classbits[32];
3059    
3060    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3061    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3062    dynamically as we process the pattern. */
3063    
3064  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3065  BOOL class_utf8;  BOOL class_utf8;
3066  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8 = (options & PCRE_UTF8) != 0;
# Line 2652  BOOL utf8 = FALSE; Line 3072  BOOL utf8 = FALSE;
3072  uschar *utf8_char = NULL;  uschar *utf8_char = NULL;
3073  #endif  #endif
3074    
3075  #ifdef DEBUG  #ifdef PCRE_DEBUG
3076  if (lengthptr != NULL) DPRINTF((">> start branch\n"));  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3077  #endif  #endif
3078    
# Line 2706  for (;; ptr++) Line 3126  for (;; ptr++)
3126    
3127    c = *ptr;    c = *ptr;
3128    
3129      /* If we are at the end of a nested substitution, revert to the outer level
3130      string. Nesting only happens one level deep. */
3131    
3132      if (c == 0 && nestptr != NULL)
3133        {
3134        ptr = nestptr;
3135        nestptr = NULL;
3136        c = *ptr;
3137        }
3138    
3139    /* 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
3140    previous cycle of this loop. */    previous cycle of this loop. */
3141    
3142    if (lengthptr != NULL)    if (lengthptr != NULL)
3143      {      {
3144  #ifdef DEBUG  #ifdef PCRE_DEBUG
3145      if (code > cd->hwm) cd->hwm = code;                 /* High water info */      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3146  #endif  #endif
3147      if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */      if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3148        {        {
3149        *errorcodeptr = ERR52;        *errorcodeptr = ERR52;
3150        goto FAILED;        goto FAILED;
# Line 2736  for (;; ptr++) Line 3166  for (;; ptr++)
3166        goto FAILED;        goto FAILED;
3167        }        }
3168    
3169      *lengthptr += code - last_code;      *lengthptr += (int)(code - last_code);
3170      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));
3171    
3172      /* 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 2763  for (;; ptr++) Line 3193  for (;; ptr++)
3193    /* 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
3194    reference list. */    reference list. */
3195    
3196    else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3197      {      {
3198      *errorcodeptr = ERR52;      *errorcodeptr = ERR52;
3199      goto FAILED;      goto FAILED;
# Line 2811  for (;; ptr++) Line 3241  for (;; ptr++)
3241      previous_callout = NULL;      previous_callout = NULL;
3242      }      }
3243    
3244    /* In extended mode, skip white space and comments */    /* In extended mode, skip white space and comments. */
3245    
3246    if ((options & PCRE_EXTENDED) != 0)    if ((options & PCRE_EXTENDED) != 0)
3247      {      {
3248      if ((cd->ctypes[c] & ctype_space) != 0) continue;      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3249      if (c == CHAR_NUMBER_SIGN)      if (c == CHAR_NUMBER_SIGN)
3250        {        {
3251        while (*(++ptr) != 0)        ptr++;
3252          while (*ptr != 0)
3253          {          {
3254          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3255            ptr++;
3256    #ifdef SUPPORT_UTF8
3257            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3258    #endif
3259          }          }
3260        if (*ptr != 0) continue;        if (*ptr != 0) continue;
3261    
# Line 2854  for (;; ptr++) Line 3289  for (;; ptr++)
3289          *errorcodeptr = ERR20;          *errorcodeptr = ERR20;
3290          goto FAILED;          goto FAILED;
3291          }          }
3292        *lengthptr += code - last_code;   /* To include callout length */        *lengthptr += (int)(code - last_code);   /* To include callout length */
3293        DPRINTF((">> end branch\n"));        DPRINTF((">> end branch\n"));
3294        }        }
3295      return TRUE;      return TRUE;
# Line 2865  for (;; ptr++) Line 3300  for (;; ptr++)
3300      the setting of any following char as a first character. */      the setting of any following char as a first character. */
3301    
3302      case CHAR_CIRCUMFLEX_ACCENT:      case CHAR_CIRCUMFLEX_ACCENT:
3303        previous = NULL;
3304      if ((options & PCRE_MULTILINE) != 0)      if ((options & PCRE_MULTILINE) != 0)
3305        {        {
3306        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3307          *code++ = OP_CIRCM;
3308        }        }
3309      previous = NULL;      else *code++ = OP_CIRC;
     *code++ = OP_CIRC;  
3310      break;      break;
3311    
3312      case CHAR_DOLLAR_SIGN:      case CHAR_DOLLAR_SIGN:
3313      previous = NULL;      previous = NULL;
3314      *code++ = OP_DOLL;      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3315      break;      break;
3316    
3317      /* 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 3059  for (;; ptr++) Line 3495  for (;; ptr++)
3495            ptr++;            ptr++;
3496            }            }
3497    
3498          posix_class = check_posix_name(ptr, tempptr - ptr);          posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3499          if (posix_class < 0)          if (posix_class < 0)
3500            {            {
3501            *errorcodeptr = ERR30;            *errorcodeptr = ERR30;
# Line 3073  for (;; ptr++) Line 3509  for (;; ptr++)
3509          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3510            posix_class = 0;            posix_class = 0;
3511    
3512          /* 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
3513          because we may be adding and subtracting from it, and we don't want to          different escape sequences that use Unicode properties. */
3514          subtract bits that may be in the main map already. At the end we or the  
3515          result into the bit map that is being built. */  #ifdef SUPPORT_UCP
3516            if ((options & PCRE_UCP) != 0)
3517              {
3518              int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3519              if (posix_substitutes[pc] != NULL)
3520                {
3521                nestptr = tempptr + 1;
3522                ptr = posix_substitutes[pc] - 1;
3523                continue;
3524                }
3525              }
3526    #endif
3527            /* In the non-UCP case, we build the bit map for the POSIX class in a
3528            chunk of local store because we may be adding and subtracting from it,
3529            and we don't want to subtract bits that may be in the main map already.
3530            At the end we or the result into the bit map that is being built. */
3531    
3532          posix_class *= 3;          posix_class *= 3;
3533    
# Line 3120  for (;; ptr++) Line 3571  for (;; ptr++)
3571    
3572        /* Backslash may introduce a single character, or it may introduce one        /* Backslash may introduce a single character, or it may introduce one
3573        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
3574        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
3575        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
3576        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
3577        character in them, so set class_charcount bigger than one. */        are either treated as literal characters (by default), or are faulted if
3578          PCRE_EXTRA is set. */
3579    
3580        if (c == CHAR_BACKSLASH)        if (c == CHAR_BACKSLASH)
3581          {          {
3582          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3583          if (*errorcodeptr != 0) goto FAILED;          if (*errorcodeptr != 0) goto FAILED;
3584    
3585          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 */  
3586          else if (-c == ESC_Q)            /* Handle start of quoted string */          else if (-c == ESC_Q)            /* Handle start of quoted string */
3587            {            {
3588            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
# Line 3149  for (;; ptr++) Line 3599  for (;; ptr++)
3599            register const uschar *cbits = cd->cbits;            register const uschar *cbits = cd->cbits;
3600            class_charcount += 2;     /* Greater than 1 is what matters */            class_charcount += 2;     /* Greater than 1 is what matters */
3601    
3602            /* Save time by not doing this in the pre-compile phase. */            switch (-c)
   
           if (lengthptr == NULL) switch (-c)  
3603              {              {
3604    #ifdef SUPPORT_UCP
3605                case ESC_du:     /* These are the values given for \d etc */
3606                case ESC_DU:     /* when PCRE_UCP is set. We replace the */
3607                case ESC_wu:     /* escape sequence with an appropriate \p */
3608                case ESC_WU:     /* or \P to test Unicode properties instead */
3609                case ESC_su:     /* of the default ASCII testing. */
3610                case ESC_SU:
3611                nestptr = ptr;
3612                ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
3613                class_charcount -= 2;                /* Undo! */
3614                continue;
3615    #endif
3616              case ESC_d:              case ESC_d:
3617              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3618              continue;              continue;
# Line 3171  for (;; ptr++) Line 3631  for (;; ptr++)
3631              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3632              continue;              continue;
3633    
3634                /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3635                if it was previously set by something earlier in the character
3636                class. */
3637    
3638              case ESC_s:              case ESC_s:
3639              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];              classbits[0] |= cbits[cbit_space];
3640              classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= cbits[cbit_space+1] & ~0x08;
3641                for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3642              continue;              continue;
3643    
3644              case ESC_S:              case ESC_S:
# Line 3182  for (;; ptr++) Line 3647  for (;; ptr++)
3647              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */
3648              continue;              continue;
3649    
3650              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)  
             {  
3651              SETBIT(classbits, 0x09); /* VT */              SETBIT(classbits, 0x09); /* VT */
3652              SETBIT(classbits, 0x20); /* SPACE */              SETBIT(classbits, 0x20); /* SPACE */
3653              SETBIT(classbits, 0xa0); /* NSBP */              SETBIT(classbits, 0xa0); /* NSBP */
# Line 3219  for (;; ptr++) Line 3671  for (;; ptr++)
3671                }                }
3672  #endif  #endif
3673              continue;              continue;
             }  
3674    
3675            if (-c == ESC_H)              case ESC_H:
             {  
3676              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3677                {                {
3678                int x = 0xff;                int x = 0xff;
# Line 3264  for (;; ptr++) Line 3714  for (;; ptr++)
3714                }                }
3715  #endif  #endif
3716              continue;              continue;
             }  
3717    
3718            if (-c == ESC_v)              case ESC_v:
             {  
3719              SETBIT(classbits, 0x0a); /* LF */              SETBIT(classbits, 0x0a); /* LF */
3720              SETBIT(classbits, 0x0b); /* VT */              SETBIT(classbits, 0x0b); /* VT */
3721              SETBIT(classbits, 0x0c); /* FF */              SETBIT(classbits, 0x0c); /* FF */
# Line 3283  for (;; ptr++) Line 3731  for (;; ptr++)
3731                }                }
3732  #endif  #endif
3733              continue;              continue;
             }  
3734    
3735            if (-c == ESC_V)              case ESC_V:
             {  
3736              for (c = 0; c < 32; c++)              for (c = 0; c < 32; c++)
3737                {                {
3738                int x = 0xff;                int x = 0xff;
# Line 3316  for (;; ptr++) Line 3762  for (;; ptr++)
3762                }                }
3763  #endif  #endif
3764              continue;              continue;
             }  
   
           /* We need to deal with \P and \p in both phases. */  
3765    
3766  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3767            if (-c == ESC_p || -c == ESC_P)              case ESC_p:
3768              {              case ESC_P:
3769              BOOL negated;                {
3770              int pdata;                BOOL negated;
3771              int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);                int pdata;
3772              if (ptype < 0) goto FAILED;                int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3773              class_utf8 = TRUE;                if (ptype < 0) goto FAILED;
3774              *class_utf8data++ = ((-c == ESC_p) != negated)?                class_utf8 = TRUE;
3775                XCL_PROP : XCL_NOTPROP;                *class_utf8data++ = ((-c == ESC_p) != negated)?
3776              *class_utf8data++ = ptype;                  XCL_PROP : XCL_NOTPROP;
3777              *class_utf8data++ = pdata;                *class_utf8data++ = ptype;
3778              class_charcount -= 2;   /* Not a < 256 character */                *class_utf8data++ = pdata;
3779              continue;                class_charcount -= 2;   /* Not a < 256 character */
3780              }                continue;
3781                  }
3782  #endif  #endif
3783            /* Unrecognized escapes are faulted if PCRE is running in its              /* Unrecognized escapes are faulted if PCRE is running in its
3784            strict mode. By default, for compatibility with Perl, they are              strict mode. By default, for compatibility with Perl, they are
3785            treated as literals. */              treated as literals. */
3786    
3787            if ((options & PCRE_EXTRA) != 0)              default:
3788              {              if ((options & PCRE_EXTRA) != 0)
3789              *errorcodeptr = ERR7;                {
3790              goto FAILED;                *errorcodeptr = ERR7;
3791                  goto FAILED;
3792                  }
3793                class_charcount -= 2;  /* Undo the default count from above */
3794                c = *ptr;              /* Get the final character and fall through */
3795                break;
3796              }              }
   
           class_charcount -= 2;  /* Undo the default count from above */  
           c = *ptr;              /* Get the final character and fall through */  
3797            }            }
3798    
3799          /* 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 3417  for (;; ptr++) Line 3863  for (;; ptr++)
3863            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);            d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3864            if (*errorcodeptr != 0) goto FAILED;            if (*errorcodeptr != 0) goto FAILED;
3865    
3866            /* \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 */  
3867    
3868            if (d < 0)            if (d < 0)
3869              {              {
3870              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  
3871                {                {
3872                ptr = oldptr;                ptr = oldptr;
3873                goto LONE_SINGLE_CHARACTER;  /* A few lines below */                goto LONE_SINGLE_CHARACTER;  /* A few lines below */
# Line 3590  for (;; ptr++) Line 4033  for (;; ptr++)
4033          }          }
4034        }        }
4035    
4036      /* 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.
4037        If we are at the end of an internal nested string, revert to the outer
4038        string. */
4039    
4040        while (((c = *(++ptr)) != 0 ||
4041               (nestptr != NULL &&
4042                 (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
4043               (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
4044    
4045      while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));      /* Check for missing terminating ']' */
4046    
4047      if (c == 0)                          /* Missing terminating ']' */      if (c == 0)
4048        {        {
4049        *errorcodeptr = ERR6;        *errorcodeptr = ERR6;
4050        goto FAILED;        goto FAILED;
4051        }        }
4052    
   
 /* 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  
   
   
4053      /* 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
4054      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
4055      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 3626  we set the flag only if there is a liter Line 4057  we set the flag only if there is a liter
4057    
4058      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
4059      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR      characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
4060      operate on single-bytes only. This is an historical hangover. Maybe one day      operate on single-bytes characters only. This is an historical hangover.
4061      we can tidy these opcodes to handle multi-byte characters.      Maybe one day we can tidy these opcodes to handle multi-byte characters.
4062    
4063      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
4064      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.
4065      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
4066      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
4067      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
4068      reqbyte, save the previous value for reinstating. */      of reqbyte, save the previous value for reinstating. */
4069    
4070  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4071      if (class_charcount == 1 && !class_utf8 &&      if (class_charcount == 1 && !class_utf8 &&
# Line 3645  we set the flag only if there is a liter Line 4076  we set the flag only if there is a liter
4076        {        {
4077        zeroreqbyte = reqbyte;        zeroreqbyte = reqbyte;
4078    
4079        /* The OP_NOT opcode works on one-byte characters only. */        /* The OP_NOT[I] opcodes work on one-byte characters only. */
4080    
4081        if (negate_class)        if (negate_class)
4082          {          {
4083          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4084          zerofirstbyte = firstbyte;          zerofirstbyte = firstbyte;
4085          *code++ = OP_NOT;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT;
4086          *code++ = class_lastchar;          *code++ = class_lastchar;
4087          break;          break;
4088          }          }
# Line 3682  we set the flag only if there is a liter Line 4113  we set the flag only if there is a liter
4113    
4114      /* If there are characters with values > 255, we have to compile an      /* If there are characters with values > 255, we have to compile an
4115      extended class, with its own opcode, unless there was a negated special      extended class, with its own opcode, unless there was a negated special
4116      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
4117      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
4118      (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
4119      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
4120        actual compiled code. */
4121    
4122  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4123      if (class_utf8 && !should_flip_negation)      if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
4124        {        {
4125        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */
4126        *code++ = OP_XCLASS;        *code++ = OP_XCLASS;
# Line 3714  we set the flag only if there is a liter Line 4146  we set the flag only if there is a liter
4146        }        }
4147  #endif  #endif
4148    
4149      /* 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
4150      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
4151      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
4152      map into the code vector, negating it if necessary. */      (non-UCP) in the class. Then copy the 32-byte map into the code vector,
4153        negating it if necessary. */
4154    
4155      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;      *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4156      if (negate_class)      if (negate_class)
# Line 3777  we set the flag only if there is a liter Line 4210  we set the flag only if there is a liter
4210      op_type = 0;                    /* Default single-char op codes */      op_type = 0;                    /* Default single-char op codes */
4211      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      possessive_quantifier = FALSE;  /* Default not possessive quantifier */
4212    
4213      /* 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
4214      for an inserted OP_ONCE for the additional '+' extension. */      insert something before it. */
4215    
4216      tempcode = previous;      tempcode = previous;
4217    
# Line 3800  we set the flag only if there is a liter Line 4233  we set the flag only if there is a liter
4233        ptr++;        ptr++;
4234        }        }
4235      else repeat_type = greedy_default;      else repeat_type = greedy_default;
4236    
4237        /* If previous was a recursion call, wrap it in atomic brackets so that
4238        previous becomes the atomic group. All recursions were so wrapped in the
4239        past, but it no longer happens for non-repeated recursions. In fact, the
4240        repeated ones could be re-implemented independently so as not to need this,
4241        but for the moment we rely on the code for repeating groups. */
4242    
4243        if (*previous == OP_RECURSE)
4244          {
4245          memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);
4246          *previous = OP_ONCE;
4247          PUT(previous, 1, 2 + 2*LINK_SIZE);
4248          previous[2 + 2*LINK_SIZE] = OP_KET;
4249          PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE);
4250          code += 2 + 2 * LINK_SIZE;
4251          length_prevgroup = 3 + 3*LINK_SIZE;
4252    
4253          /* When actually compiling, we need to check whether this was a forward
4254          reference, and if so, adjust the offset. */
4255    
4256          if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE)
4257            {
4258            int offset = GET(cd->hwm, -LINK_SIZE);
4259            if (offset == previous + 1 - cd->start_code)
4260              PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE);
4261            }
4262          }
4263    
4264        /* Now handle repetition for the different types of item. */
4265    
4266      /* If previous was a character match, abolish the item and generate a      /* If previous was a character match, abolish the item and generate a
4267      repeat item instead. If a char item has a minumum of more than one, ensure      repeat item instead. If a char item has a minumum of more than one, ensure
# Line 3807  we set the flag only if there is a liter Line 4269  we set the flag only if there is a liter
4269      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
4270      instead.  */      instead.  */
4271    
4272      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (*previous == OP_CHAR || *previous == OP_CHARI)
4273        {        {
4274          op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR;
4275    
4276        /* 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
4277        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
4278        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 3841  we set the flag only if there is a liter Line 4305  we set the flag only if there is a liter
4305    
4306        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4307            repeat_max < 0 &&            repeat_max < 0 &&
4308            check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
             options, cd))  
4309          {          {
4310          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4311          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3854  we set the flag only if there is a liter Line 4317  we set the flag only if there is a liter
4317      /* If previous was a single negated character ([^a] or similar), we use      /* If previous was a single negated character ([^a] or similar), we use
4318      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-
4319      character repeats by setting opt_type to add a suitable offset into      character repeats by setting opt_type to add a suitable offset into
4320      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
4321      currently used only for single-byte chars. */      are currently used only for single-byte chars. */
4322    
4323      else if (*previous == OP_NOT)      else if (*previous == OP_NOT || *previous == OP_NOTI)
4324        {        {
4325        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR;
4326        c = previous[1];        c = previous[1];
4327        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4328            repeat_max < 0 &&            repeat_max < 0 &&
4329            check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4330          {          {
4331          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4332          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3887  we set the flag only if there is a liter Line 4350  we set the flag only if there is a liter
4350    
4351        if (!possessive_quantifier &&        if (!possessive_quantifier &&
4352            repeat_max < 0 &&            repeat_max < 0 &&
4353            check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))            check_auto_possessive(previous, utf8, ptr + 1, options, cd))
4354          {          {
4355          repeat_type = 0;    /* Force greedy */          repeat_type = 0;    /* Force greedy */
4356          possessive_quantifier = TRUE;          possessive_quantifier = TRUE;
# Line 3909  we set the flag only if there is a liter Line 4372  we set the flag only if there is a liter
4372    
4373        if (repeat_max == 0) goto END_REPEAT;        if (repeat_max == 0) goto END_REPEAT;
4374    
4375        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4376        /* This code is obsolete from release 8.00; the restriction was finally        /* This code is obsolete from release 8.00; the restriction was finally
4377        removed: */        removed: */
4378    
4379        /* All real repeats make it impossible to handle partial matching (maybe        /* All real repeats make it impossible to handle partial matching (maybe
4380        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4381    
4382        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4383        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4384    
4385        /* Combine the op_type with the repeat_type */        /* Combine the op_type with the repeat_type */
4386    
# Line 4056  we set the flag only if there is a liter Line 4519  we set the flag only if there is a liter
4519  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
4520               *previous == OP_XCLASS ||               *previous == OP_XCLASS ||
4521  #endif  #endif
4522               *previous == OP_REF)               *previous == OP_REF ||
4523                 *previous == OP_REFI)
4524        {        {
4525        if (repeat_max == 0)        if (repeat_max == 0)
4526          {          {
# Line 4064  we set the flag only if there is a liter Line 4528  we set the flag only if there is a liter
4528          goto END_REPEAT;          goto END_REPEAT;
4529          }          }
4530    
4531        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4532        /* This code is obsolete from release 8.00; the restriction was finally        /* This code is obsolete from release 8.00; the restriction was finally
4533        removed: */        removed: */
4534    
# Line 4072  we set the flag only if there is a liter Line 4536  we set the flag only if there is a liter
4536        one day we will be able to remove this restriction). */        one day we will be able to remove this restriction). */
4537    
4538        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */        /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4539        /*--------------------------------------------------------------------*/        /*--------------------------------------------------------------------*/
4540    
4541        if (repeat_min == 0 && repeat_max == -1)        if (repeat_min == 0 && repeat_max == -1)
4542          *code++ = OP_CRSTAR + repeat_type;          *code++ = OP_CRSTAR + repeat_type;
# Line 4090  we set the flag only if there is a liter Line 4554  we set the flag only if there is a liter
4554        }        }
4555    
4556      /* 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
4557      cases. */      cases. Note that at this point we can encounter only the "basic" bracket
4558        opcodes such as BRA and CBRA, as this is the place where they get converted
4559        into the more special varieties such as BRAPOS and SBRA. A test for >=
4560        OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK,
4561        ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow
4562        repetition of assertions, but now it does, for Perl compatibility. */
4563    
4564      else if (*previous == OP_BRA  || *previous == OP_CBRA ||      else if (*previous >= OP_ASSERT && *previous <= OP_COND)
              *previous == OP_ONCE || *previous == OP_COND)  
4565        {        {
4566        register int i;        register int i;
4567        int ketoffset = 0;        int len = (int)(code - previous);
       int len = code - previous;  
4568        uschar *bralink = NULL;        uschar *bralink = NULL;
4569          uschar *brazeroptr = NULL;
4570        /* Repeating a DEFINE group is pointless */  
4571          /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so
4572          we just ignore the repeat. */
4573    
4574        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)        if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4575          {          goto END_REPEAT;
         *errorcodeptr = ERR55;  
         goto FAILED;  
         }  
4576    
4577        /* If the maximum repeat count is unlimited, find the end of the bracket        /* There is no sense in actually repeating assertions. The only potential
4578        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,
4579        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
4580        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. */
4581        pointer. */  
4582          if (*previous < OP_ONCE)    /* Assertion */
4583        if (repeat_max == -1)          {
4584          {          if (repeat_min > 0) goto END_REPEAT;
4585          register uschar *ket = previous;          if (repeat_max < 0 || repeat_max > 1) repeat_max = 1;
4586          do ket += GET(ket, 1); while (*ket != OP_KET);          }
         ketoffset = code - ket;  
         }  
4587    
4588        /* 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
4589        OP_BRAZERO in front of it, and because the group appears once in the        OP_BRAZERO in front of it, and because the group appears once in the
# Line 4139  we set the flag only if there is a liter Line 4603  we set the flag only if there is a liter
4603          **   goto END_REPEAT;          **   goto END_REPEAT;
4604          **   }          **   }
4605    
4606          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
4607          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
4608          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
4609          groups are referenced, we cannot do this selectively.          don't have a list of which groups are referenced, we cannot do this
4610            selectively.
4611    
4612          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
4613          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 4162  we set the flag only if there is a liter Line 4627  we set the flag only if there is a liter
4627              *previous++ = OP_SKIPZERO;              *previous++ = OP_SKIPZERO;
4628              goto END_REPEAT;              goto END_REPEAT;
4629              }              }
4630              brazeroptr = previous;    /* Save for possessive optimizing */
4631            *previous++ = OP_BRAZERO + repeat_type;            *previous++ = OP_BRAZERO + repeat_type;
4632            }            }
4633    
# Line 4186  we set the flag only if there is a liter Line 4652  we set the flag only if there is a liter
4652            /* We chain together the bracket offset fields that have to be            /* We chain together the bracket offset fields that have to be
4653            filled in later when the ends of the brackets are reached. */            filled in later when the ends of the brackets are reached. */
4654    
4655            offset = (bralink == NULL)? 0 : previous - bralink;            offset = (bralink == NULL)? 0 : (int)(previous - bralink);
4656            bralink = previous;            bralink = previous;
4657            PUTINC(previous, 0, offset);            PUTINC(previous, 0, offset);
4658            }            }
# Line 4207  we set the flag only if there is a liter Line 4673  we set the flag only if there is a liter
4673            {            {
4674            /* 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
4675            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
4676            potential integer overflow. */            potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4677              integer type when available, otherwise double. */
4678    
4679            if (lengthptr != NULL)            if (lengthptr != NULL)
4680              {              {
4681              int delta = (repeat_min - 1)*length_prevgroup;              int delta = (repeat_min - 1)*length_prevgroup;
4682              if ((double)(repeat_min - 1)*(double)length_prevgroup >              if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4683                                                              (double)INT_MAX ||                    (INT64_OR_DOUBLE)length_prevgroup >
4684                        (INT64_OR_DOUBLE)INT_MAX ||
4685                  OFLOW_MAX - *lengthptr < delta)                  OFLOW_MAX - *lengthptr < delta)
4686                {                {
4687                *errorcodeptr = ERR20;                *errorcodeptr = ERR20;
# Line 4259  we set the flag only if there is a liter Line 4727  we set the flag only if there is a liter
4727          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
4728          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
4729          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
4730          paranoid checks to avoid integer overflow. */          paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4731            a 64-bit integer type when available, otherwise double. */
4732    
4733          if (lengthptr != NULL && repeat_max > 0)          if (lengthptr != NULL && repeat_max > 0)
4734            {            {
4735            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -            int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4736                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */                        2 - 2*LINK_SIZE;   /* Last one doesn't nest */
4737            if ((double)repeat_max *            if ((INT64_OR_DOUBLE)repeat_max *
4738                  (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)                  (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4739                    > (double)INT_MAX ||                    > (INT64_OR_DOUBLE)INT_MAX ||
4740                OFLOW_MAX - *lengthptr < delta)                OFLOW_MAX - *lengthptr < delta)
4741              {              {
4742              *errorcodeptr = ERR20;              *errorcodeptr = ERR20;
# Line 4292  we set the flag only if there is a liter Line 4761  we set the flag only if there is a liter
4761              {              {
4762              int offset;              int offset;
4763              *code++ = OP_BRA;              *code++ = OP_BRA;
4764              offset = (bralink == NULL)? 0 : code - bralink;              offset = (bralink == NULL)? 0 : (int)(code - bralink);
4765              bralink = code;              bralink = code;
4766              PUTINC(code, 0, offset);              PUTINC(code, 0, offset);
4767              }              }
# Line 4313  we set the flag only if there is a liter Line 4782  we set the flag only if there is a liter
4782          while (bralink != NULL)          while (bralink != NULL)
4783            {            {
4784            int oldlinkoffset;            int oldlinkoffset;
4785            int offset = code - bralink + 1;            int offset = (int)(code - bralink + 1);
4786            uschar *bra = code - offset;            uschar *bra = code - offset;
4787            oldlinkoffset = GET(bra, 1);            oldlinkoffset = GET(bra, 1);
4788            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;            bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
# Line 4323  we set the flag only if there is a liter Line 4792  we set the flag only if there is a liter
4792            }            }
4793          }          }
4794    
4795        /* 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
4796        can't just offset backwards from the current code point, because we        ONCE brackets, that's all we need to do. However, possessively repeated
4797        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
4798        correct offset was computed above.        behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to
4799          deal with possessive ONCEs specially.
4800    
4801          Otherwise, if the quantifier was possessive, we convert the BRA code to
4802          the POS form, and the KET code to KETRPOS. (It turns out to be convenient
4803          at runtime to detect this kind of subpattern at both the start and at the
4804          end.) The use of special opcodes makes it possible to reduce greatly the
4805          stack usage in pcre_exec(). If the group is preceded by OP_BRAZERO,
4806          convert this to OP_BRAPOSZERO. Then cancel the possessive flag so that
4807          the default action below, of wrapping everything inside atomic brackets,
4808          does not happen.
4809    
4810        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
4811        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
4812        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
4813        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
4814        atomic groups at runtime, but in a different way.] */        runtime, but in a different way.] */
4815    
4816        else        else
4817          {          {
4818          uschar *ketcode = code - ketoffset;          uschar *ketcode = code - 1 - LINK_SIZE;
4819          uschar *bracode = ketcode - GET(ketcode, 1);          uschar *bracode = ketcode - GET(ketcode, 1);
4820          *ketcode = OP_KETRMAX + repeat_type;  
4821          if (lengthptr == NULL && *bracode != OP_ONCE)          if (*bracode == OP_ONCE && possessive_quantifier) *bracode = OP_BRA;
4822            if (*bracode == OP_ONCE)
4823              *ketcode = OP_KETRMAX + repeat_type;
4824            else
4825            {            {
4826            uschar *scode = bracode;            if (possessive_quantifier)
           do  
4827              {              {
4828              if (could_be_empty_branch(scode, ketcode, utf8))              *bracode += 1;                   /* Switch to xxxPOS opcodes */
4829                *ketcode = OP_KETRPOS;
4830                if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO;
4831                possessive_quantifier = FALSE;
4832                }
4833              else *ketcode = OP_KETRMAX + repeat_type;
4834    
4835              if (lengthptr == NULL)
4836                {
4837                uschar *scode = bracode;
4838                do
4839                {                {
4840                *bracode += OP_SBRA - OP_BRA;                if (could_be_empty_branch(scode, ketcode, utf8, cd))
4841                break;                  {
4842                    *bracode += OP_SBRA - OP_BRA;
4843                    break;
4844                    }
4845                  scode += GET(scode, 1);
4846                }                }
4847              scode += GET(scode, 1);              while (*scode == OP_ALT);
4848              }              }
           while (*scode == OP_ALT);  
4849            }            }
4850          }          }
4851        }        }
# Line 4372  we set the flag only if there is a liter Line 4866  we set the flag only if there is a liter
4866        }        }
4867    
4868      /* If the character following a repeat is '+', or if certain optimization      /* If the character following a repeat is '+', or if certain optimization
4869      tests above succeeded, possessive_quantifier is TRUE. For some of the      tests above succeeded, possessive_quantifier is TRUE. For some opcodes,
4870      simpler opcodes, there is an special alternative opcode for this. For      there are special alternative opcodes for this case. For anything else, we
4871      anything else, we wrap the entire repeated item inside OP_ONCE brackets.      wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+'
4872      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
4873      but the special opcodes can optimize it a bit. The repeated item starts at      special opcodes can optimize it.
4874      tempcode, not at previous, which might be the first part of a string whose  
4875      (former) last char we repeated.      Possessively repeated subpatterns have already been handled in the code
4876        just above, so possessive_quantifier is always FALSE for them at this
4877        stage.
4878    
4879        Note that the repeated item starts at tempcode, not at previous, which
4880        might be the first part of a string whose (former) last char we repeated.
4881    
4882      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
4883      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 4387  we set the flag only if there is a liter Line 4886  we set the flag only if there is a liter
4886      if (possessive_quantifier)      if (possessive_quantifier)
4887        {        {
4888        int len;        int len;
4889    
4890        if (*tempcode == OP_TYPEEXACT)        if (*tempcode == OP_TYPEEXACT)
4891          tempcode += _pcre_OP_lengths[*tempcode] +          tempcode += _pcre_OP_lengths[*tempcode] +
4892            ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);            ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4893    
4894        else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)        else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4895          {          {
4896          tempcode += _pcre_OP_lengths[*tempcode];          tempcode += _pcre_OP_lengths[*tempcode];
# Line 4399  we set the flag only if there is a liter Line 4898  we set the flag only if there is a liter
4898          if (utf8 && tempcode[-1] >= 0xc0)          if (utf8 && tempcode[-1] >= 0xc0)
4899            tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];            tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4900  #endif  #endif
4901          }          }
4902    
4903        len = code - tempcode;        len = (int)(code - tempcode);
4904        if (len > 0) switch (*tempcode)        if (len > 0) switch (*tempcode)
4905          {          {
4906          case OP_STAR:  *tempcode = OP_POSSTAR; break;          case OP_STAR:  *tempcode = OP_POSSTAR; break;
# Line 4409  we set the flag only if there is a liter Line 4908  we set the flag only if there is a liter
4908          case OP_QUERY: *tempcode = OP_POSQUERY; break;          case OP_QUERY: *tempcode = OP_POSQUERY; break;
4909          case OP_UPTO:  *tempcode = OP_POSUPTO; break;          case OP_UPTO:  *tempcode = OP_POSUPTO; break;
4910    
4911          case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;          case OP_STARI:  *tempcode = OP_POSSTARI; break;
4912          case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;          case OP_PLUSI:  *tempcode = OP_POSPLUSI; break;
4913          case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;          case OP_QUERYI: *tempcode = OP_POSQUERYI; break;
4914          case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;          case OP_UPTOI:  *tempcode = OP_POSUPTOI; break;
4915    
4916          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;          case OP_NOTSTAR:  *tempcode = OP_NOTPOSSTAR; break;
4917          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;          case OP_NOTPLUS:  *tempcode = OP_NOTPOSPLUS; break;
4918          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;          case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4919          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;          case OP_NOTUPTO:  *tempcode = OP_NOTPOSUPTO; break;
4920    
4921            case OP_NOTSTARI:  *tempcode = OP_NOTPOSSTARI; break;
4922            case OP_NOTPLUSI:  *tempcode = OP_NOTPOSPLUSI; break;
4923            case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break;
4924            case OP_NOTUPTOI:  *tempcode = OP_NOTPOSUPTOI; break;
4925    
4926            case OP_TYPESTAR:  *tempcode = OP_TYPEPOSSTAR; break;
4927            case OP_TYPEPLUS:  *tempcode = OP_TYPEPOSPLUS; break;
4928            case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4929            case OP_TYPEUPTO:  *tempcode = OP_TYPEPOSUPTO; break;
4930    
4931            /* Because we are moving code along, we must ensure that any
4932            pending recursive references are updated. */
4933    
4934          default:          default:
4935            *code = OP_END;
4936            adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4937          memmove(tempcode + 1+LINK_SIZE, tempcode, len);          memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4938          code += 1 + LINK_SIZE;          code += 1 + LINK_SIZE;
4939          len += 1 + LINK_SIZE;          len += 1 + LINK_SIZE;
# Line 4455  we set the flag only if there is a liter Line 4969  we set the flag only if there is a liter
4969    
4970      /* First deal with various "verbs" that can be introduced by '*'. */      /* First deal with various "verbs" that can be introduced by '*'. */
4971    
4972      if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)      if (*(++ptr) == CHAR_ASTERISK &&
4973             ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
4974        {        {
4975        int i, namelen;        int i, namelen;
4976          int arglen = 0;
4977        const char *vn = verbnames;        const char *vn = verbnames;
4978        const uschar *name = ++ptr;        const uschar *name = ptr + 1;
4979          const uschar *arg = NULL;
4980        previous = NULL;        previous = NULL;
4981        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};        while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4982          namelen = (int)(ptr - name);
4983    
4984        if (*ptr == CHAR_COLON)        if (*ptr == CHAR_COLON)
4985          {          {
4986          *errorcodeptr = ERR59;   /* Not supported */          arg = ++ptr;
4987          goto FAILED;          while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
4988              || *ptr == '_') ptr++;
4989            arglen = (int)(ptr - arg);
4990          }          }
4991    
4992        if (*ptr != CHAR_RIGHT_PARENTHESIS)        if (*ptr != CHAR_RIGHT_PARENTHESIS)
4993          {          {
4994          *errorcodeptr = ERR60;          *errorcodeptr = ERR60;
4995          goto FAILED;          goto FAILED;
4996          }          }
4997        namelen = ptr - name;  
4998          /* Scan the table of verb names */
4999    
5000        for (i = 0; i < verbcount; i++)        for (i = 0; i < verbcount; i++)
5001          {          {
5002          if (namelen == verbs[i].len &&          if (namelen == verbs[i].len &&
5003              strncmp((char *)name, vn, namelen) == 0)              strncmp((char *)name, vn, namelen) == 0)
5004            {            {
5005            /* Check for open captures before ACCEPT */            /* Check for open captures before ACCEPT and convert it to
5006              ASSERT_ACCEPT if in an assertion. */
5007    
5008            if (verbs[i].op == OP_ACCEPT)            if (verbs[i].op == OP_ACCEPT)
5009              {              {
5010              open_capitem *oc;              open_capitem *oc;
5011              cd->had_accept = TRUE;              if (arglen != 0)
5012                  {
5013                  *errorcodeptr = ERR59;
5014                  goto FAILED;
5015                  }
5016                cd->had_accept = TRUE;
5017              for (oc = cd->open_caps; oc != NULL; oc = oc->next)              for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5018                {                {
5019                *code++ = OP_CLOSE;                *code++ = OP_CLOSE;
5020                PUT2INC(code, 0, oc->number);                PUT2INC(code, 0, oc->number);
5021                }                }
5022              }              *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT;
5023            *code++ = verbs[i].op;              }
5024            break;  
5025              /* Handle other cases with/without an argument */
5026    
5027              else if (arglen == 0)
5028                {
5029                if (verbs[i].op < 0)   /* Argument is mandatory */
5030                  {
5031                  *errorcodeptr = ERR66;
5032                  goto FAILED;
5033                  }
5034                *code = verbs[i].op;
5035                if (*code++ == OP_THEN)
5036                  {
5037                  PUT(code, 0, code - bcptr->current_branch - 1);
5038                  code += LINK_SIZE;
5039                  }
5040                }
5041    
5042              else
5043                {
5044                if (verbs[i].op_arg < 0)   /* Argument is forbidden */
5045                  {
5046                  *errorcodeptr = ERR59;
5047                  goto FAILED;
5048                  }
5049                *code = verbs[i].op_arg;
5050                if (*code++ == OP_THEN_ARG)
5051                  {
5052                  PUT(code, 0, code - bcptr->current_branch - 1);
5053                  code += LINK_SIZE;
5054                  }
5055                *code++ = arglen;
5056                memcpy(code, arg, arglen);
5057                code += arglen;
5058                *code++ = 0;
5059                }
5060    
5061              break;  /* Found verb, exit loop */
5062            }            }
5063    
5064          vn += verbs[i].len + 1;          vn += verbs[i].len + 1;
5065          }          }
5066        if (i < verbcount) continue;  
5067        *errorcodeptr = ERR60;        if (i < verbcount) continue;    /* Successfully handled a verb */
5068          *errorcodeptr = ERR60;          /* Verb not recognized */
5069        goto FAILED;        goto FAILED;
5070        }        }
5071    
# Line 4615  we set the flag only if there is a liter Line 5184  we set the flag only if there is a liter
5184                recno * 10 + *ptr - CHAR_0 : -1;                recno * 10 + *ptr - CHAR_0 : -1;
5185            ptr++;            ptr++;
5186            }            }
5187          namelen = ptr - name;          namelen = (int)(ptr - name);
5188    
5189          if ((terminator > 0 && *ptr++ != terminator) ||          if ((terminator > 0 && *ptr++ != terminator) ||
5190              *ptr++ != CHAR_RIGHT_PARENTHESIS)              *ptr++ != CHAR_RIGHT_PARENTHESIS)
# Line 4652  we set the flag only if there is a liter Line 5221  we set the flag only if there is a liter
5221            }            }
5222    
5223          /* Otherwise (did not start with "+" or "-"), start by looking for the          /* Otherwise (did not start with "+" or "-"), start by looking for the
5224          name. */          name. If we find a name, add one to the opcode to change OP_CREF or
5225            OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
5226            except they record that the reference was originally to a name. The
5227            information is used to check duplicate names. */
5228    
5229          slot = cd->name_table;          slot = cd->name_table;
5230          for (i = 0; i < cd->names_found; i++)          for (i = 0; i < cd->names_found; i++)
# Line 4667  we set the flag only if there is a liter Line 5239  we set the flag only if there is a liter
5239            {            {
5240            recno = GET2(slot, 0);            recno = GET2(slot, 0);
5241            PUT2(code, 2+LINK_SIZE, recno);            PUT2(code, 2+LINK_SIZE, recno);
5242              code[1+LINK_SIZE]++;
5243            }            }
5244    
5245          /* Search the pattern for a forward reference */          /* Search the pattern for a forward reference */
5246    
5247          else if ((i = find_parens(cd, name, namelen,          else if ((i = find_parens(cd, name, namelen,
5248                          (options & PCRE_EXTENDED) != 0)) > 0)                          (options & PCRE_EXTENDED) != 0, utf8)) > 0)
5249            {            {
5250            PUT2(code, 2+LINK_SIZE, i);            PUT2(code, 2+LINK_SIZE, i);
5251              code[1+LINK_SIZE]++;
5252            }            }
5253    
5254          /* If terminator == 0 it means that the name followed directly after          /* If terminator == 0 it means that the name followed directly after
# Line 4739  we set the flag only if there is a liter Line 5313  we set the flag only if there is a liter
5313          /* ------------------------------------------------------------ */          /* ------------------------------------------------------------ */
5314          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */          case CHAR_EQUALS_SIGN:                 /* Positive lookahead */
5315          bravalue = OP_ASSERT;          bravalue = OP_ASSERT;
5316            cd->assert_depth += 1;
5317          ptr++;          ptr++;
5318          break;          break;
5319    
# Line 4753  we set the flag only if there is a liter Line 5328  we set the flag only if there is a liter
5328            continue;            continue;
5329            }            }
5330          bravalue = OP_ASSERT_NOT;          bravalue = OP_ASSERT_NOT;
5331            cd->assert_depth += 1;
5332          break;          break;
5333    
5334    
# Line 4762  we set the flag only if there is a liter Line 5338  we set the flag only if there is a liter
5338            {            {
5339            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */            case CHAR_EQUALS_SIGN:               /* Positive lookbehind */
5340            bravalue = OP_ASSERTBACK;            bravalue = OP_ASSERTBACK;
5341              cd->assert_depth += 1;
5342            ptr += 2;            ptr += 2;
5343            break;            break;
5344    
5345            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */            case CHAR_EXCLAMATION_MARK:          /* Negative lookbehind */
5346            bravalue = OP_ASSERTBACK_NOT;            bravalue = OP_ASSERTBACK_NOT;
5347              cd->assert_depth += 1;
5348            ptr += 2;            ptr += 2;
5349            break;            break;
5350    
# Line 4806  we set the flag only if there is a liter Line 5384  we set the flag only if there is a liter
5384              goto FAILED;              goto FAILED;
5385              }              }
5386            *code++ = n;            *code++ = n;
5387            PUT(code, 0, ptr - cd->start_pattern + 1);  /* Pattern offset */            PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
5388            PUT(code, LINK_SIZE, 0);                    /* Default length */            PUT(code, LINK_SIZE, 0);                          /* Default length */
5389            code += 2 * LINK_SIZE;            code += 2 * LINK_SIZE;
5390            }            }
5391          previous = NULL;          previous = NULL;
# Line 4840  we set the flag only if there is a liter Line 5418  we set the flag only if there is a liter
5418            name = ++ptr;            name = ++ptr;
5419    
5420            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;            while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5421            namelen = ptr - name;            namelen = (int)(ptr - name);
5422    
5423            /* In the pre-compile phase, just do a syntax check. */            /* In the pre-compile phase, just do a syntax check. */
5424    
# Line 4867  we set the flag only if there is a liter Line 5445  we set the flag only if there is a liter
5445                }                }
5446              }              }
5447    
5448            /* In the real compile, create the entry in the table */            /* In the real compile, create the entry in the table, maintaining
5449              alphabetical order. Duplicate names for different numbers are
5450              permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
5451              number are always OK. (An existing number can be re-used if (?|
5452              appears in the pattern.) In either event, a duplicate name results in
5453              a duplicate entry in the table, even if the number is the same. This
5454              is because the number of names, and hence the table size, is computed
5455              in the pre-compile, and it affects various numbers and pointers which
5456              would all have to be modified, and the compiled code moved down, if
5457              duplicates with the same number were omitted from the table. This
5458              doesn't seem worth the hassle. However, *different* names for the
5459              same number are not permitted. */
5460    
5461            else            else
5462              {              {
5463                BOOL dupname = FALSE;
5464              slot = cd->name_table;              slot = cd->name_table;
5465    
5466              for (i = 0; i < cd->names_found; i++)              for (i = 0; i < cd->names_found; i++)
5467                {                {
5468                int crc = memcmp(name, slot+2, namelen);                int crc = memcmp(name, slot+2, namelen);
# Line 4879  we set the flag only if there is a liter Line 5470  we set the flag only if there is a liter
5470                  {                  {
5471                  if (slot[2+namelen] == 0)                  if (slot[2+namelen] == 0)
5472                    {                    {
5473                    if ((options & PCRE_DUPNAMES) == 0)                    if (GET2(slot, 0) != cd->bracount + 1 &&
5474                          (options & PCRE_DUPNAMES) == 0)
5475                      {                      {
5476                      *errorcodeptr = ERR43;                      *errorcodeptr = ERR43;
5477                      goto FAILED;                      goto FAILED;
5478                      }                      }
5479                      else dupname = TRUE;
5480                    }                    }
5481                  else crc = -1;      /* Current name is substring */                  else crc = -1;      /* Current name is a substring */
5482                  }                  }
5483    
5484                  /* Make space in the table and break the loop for an earlier
5485                  name. For a duplicate or later name, carry on. We do this for
5486                  duplicates so that in the simple case (when ?(| is not used) they
5487                  are in order of their numbers. */
5488    
5489                if (crc < 0)                if (crc < 0)
5490                  {                  {
5491                  memmove(slot + cd->name_entry_size, slot,                  memmove(slot + cd->name_entry_size, slot,
5492                    (cd->names_found - i) * cd->name_entry_size);                    (cd->names_found - i) * cd->name_entry_size);
5493                  break;                  break;
5494                  }                  }
5495    
5496                  /* Continue the loop for a later or duplicate name */
5497    
5498                slot += cd->name_entry_size;                slot += cd->name_entry_size;
5499                }                }
5500    
5501                /* For non-duplicate names, check for a duplicate number before
5502                adding the new name. */
5503    
5504                if (!dupname)
5505                  {
5506                  uschar *cslot = cd->name_table;
5507                  for (i = 0; i < cd->names_found; i++)
5508                    {
5509                    if (cslot != slot)
5510                      {
5511                      if (GET2(cslot, 0) == cd->bracount + 1)
5512                        {
5513                        *errorcodeptr = ERR65;
5514                        goto FAILED;
5515                        }
5516                      }
5517                    else i--;
5518                    cslot += cd->name_entry_size;
5519                    }
5520                  }
5521    
5522              PUT2(slot, 0, cd->bracount + 1);              PUT2(slot, 0, cd->bracount + 1);
5523              memcpy(slot + 2, name, namelen);              memcpy(slot + 2, name, namelen);
5524              slot[2+namelen] = 0;              slot[2+namelen] = 0;
5525              }              }
5526            }            }
5527    
5528          /* In both cases, count the number of names we've encountered. */          /* In both pre-compile and compile, count the number of names we've
5529            encountered. */
5530    
         ptr++;                    /* Move past > or ' */  
5531          cd->names_found++;          cd->names_found++;
5532            ptr++;                    /* Move past > or ' */
5533          goto NUMBERED_GROUP;          goto NUMBERED_GROUP;
5534    
5535    
# Line 4924  we set the flag only if there is a liter Line 5548  we set the flag only if there is a liter
5548          NAMED_REF_OR_RECURSE:          NAMED_REF_OR_RECURSE:
5549          name = ++ptr;          name = ++ptr;
5550          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;          while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5551          namelen = ptr - name;          namelen = (int)(ptr - name);
5552    
5553          /* 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
5554          reference number. */          a dummy reference number, because it was not used in the first pass.
5555            However, with the change of recursive back references to be atomic,
5556            we have to look for the number so that this state can be identified, as
5557            otherwise the incorrect length is computed. If it's not a backwards
5558            reference, the dummy number will do. */
5559    
5560          if (lengthptr != NULL)          if (lengthptr != NULL)
5561            {            {
5562              const uschar *temp;
5563    
5564            if (namelen == 0)            if (namelen == 0)
5565              {              {
5566              *errorcodeptr = ERR62;              *errorcodeptr = ERR62;
# Line 4946  we set the flag only if there is a liter Line 5576  we set the flag only if there is a liter
5576              *errorcodeptr = ERR48;              *errorcodeptr = ERR48;
5577              goto FAILED;              goto FAILED;
5578              }              }
5579            recno = 0;  
5580              /* The name table does not exist in the first pass, so we cannot
5581              do a simple search as in the code below. Instead, we have to scan the
5582              pattern to find the number. It is important that we scan it only as
5583              far as we have got because the syntax of named subpatterns has not
5584              been checked for the rest of the pattern, and find_parens() assumes
5585              correct syntax. In any case, it's a waste of resources to scan
5586              further. We stop the scan at the current point by temporarily
5587              adjusting the value of cd->endpattern. */
5588    
5589              temp = cd->end_pattern;
5590              cd->end_pattern = ptr;
5591              recno = find_parens(cd, name, namelen,
5592                (options & PCRE_EXTENDED) != 0, utf8);
5593              cd->end_pattern = temp;
5594              if (recno < 0) recno = 0;    /* Forward ref; set dummy number */
5595            }            }
5596    
5597          /* 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 4971  we set the flag only if there is a liter Line 5616  we set the flag only if there is a liter
5616              }              }
5617            else if ((recno =                /* Forward back reference */            else if ((recno =                /* Forward back reference */
5618                      find_parens(cd, name, namelen,                      find_parens(cd, name, namelen,
5619                        (options & PCRE_EXTENDED) != 0)) <= 0)                        (options & PCRE_EXTENDED) != 0, utf8)) <= 0)
5620              {              {
5621              *errorcodeptr = ERR15;              *errorcodeptr = ERR15;
5622              goto FAILED;              goto FAILED;
# Line 5074  we set the flag only if there is a liter Line 5719  we set the flag only if there is a liter
5719            if (lengthptr == NULL)            if (lengthptr == NULL)
5720              {              {
5721              *code = OP_END;              *code = OP_END;
5722              if (recno != 0)              if (recno != 0)
5723                called = _pcre_find_bracket(cd->start_code, utf8, recno);                called = _pcre_find_bracket(cd->start_code, utf8, recno);
5724    
5725              /* Forward reference */              /* Forward reference */
# Line 5082  we set the flag only if there is a liter Line 5727  we set the flag only if there is a liter
5727              if (called == NULL)              if (called == NULL)
5728                {                {
5729                if (find_parens(cd, NULL, recno,                if (find_parens(cd, NULL, recno,
5730                      (options & PCRE_EXTENDED) != 0) < 0)                      (options & PCRE_EXTENDED) != 0, utf8) < 0)
5731                  {                  {
5732                  *errorcodeptr = ERR15;                  *errorcodeptr = ERR15;
5733                  goto FAILED;                  goto FAILED;
5734                  }                  }
5735    
5736                  /* Fudge the value of "called" so that when it is inserted as an
5737                  offset below, what it actually inserted is the reference number
5738                  of the group. Then remember the forward reference. */
5739    
5740                called = cd->start_code + recno;                called = cd->start_code + recno;
5741                PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);                PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code));
5742                }                }
5743    
5744              /* If not a forward reference, and the subpattern is still open,              /* If not a forward reference, and the subpattern is still open,
# Line 5096  we set the flag only if there is a liter Line 5746  we set the flag only if there is a liter
5746              recursion that could loop for ever, and diagnose that case. */              recursion that could loop for ever, and diagnose that case. */
5747    
5748              else if (GET(called, 1) == 0 &&              else if (GET(called, 1) == 0 &&
5749                       could_be_empty(called, code, bcptr, utf8))                       could_be_empty(called, code, bcptr, utf8, cd))
5750                {                {
5751                *errorcodeptr = ERR40;                *errorcodeptr = ERR40;
5752                goto FAILED;                goto FAILED;
5753                }                }
5754              }              }
5755    
5756            /* Insert the recursion/subroutine item, automatically wrapped inside            /* Insert the recursion/subroutine item. */
5757            "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;  
   
5758            *code = OP_RECURSE;            *code = OP_RECURSE;
5759            PUT(code, 1, called - cd->start_code);            PUT(code, 1, (int)(called - cd->start_code));
           code += 1 + LINK_SIZE;  
   
           *code = OP_KET;  
           PUT(code, 1, 2 + 2*LINK_SIZE);  
5760            code += 1 + LINK_SIZE;            code += 1 + LINK_SIZE;
   
           length_prevgroup = 3 + 3*LINK_SIZE;  
5761            }            }
5762    
5763          /* Can't determine a first byte now */          /* Can't determine a first byte now */
# Line 5180  we set the flag only if there is a liter Line 5818  we set the flag only if there is a liter
5818          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
5819          both phases.          both phases.
5820    
5821          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
5822          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. */  
5823    
5824          if (*ptr == CHAR_RIGHT_PARENTHESIS)          if (*ptr == CHAR_RIGHT_PARENTHESIS)
5825            {            {
# Line 5191  we set the flag only if there is a liter Line 5828  we set the flag only if there is a liter
5828              {              {
5829              cd->external_options = newoptions;              cd->external_options = newoptions;
5830              }              }
5831           else            else
5832              {              {
             if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))  
               {  
               *code++ = OP_OPT;  
               *code++ = newoptions & PCRE_IMS;  
               }  
5833              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);              greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5834              greedy_non_default = greedy_default ^ 1;              greedy_non_default = greedy_default ^ 1;
5835              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;              req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5836              }              }
5837    
5838            /* Change options at this level, and pass them back for use            /* Change options at this level, and pass them back for use
5839            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). */  
5840    
5841            *optionsptr = options = newoptions;            *optionsptr = options = newoptions;
5842            previous = NULL;       /* This item can't be repeated */            previous = NULL;       /* This item can't be repeated */
# Line 5223  we set the flag only if there is a liter Line 5853  we set the flag only if there is a liter
5853          }     /* End of switch for character following (? */          }     /* End of switch for character following (? */
5854        }       /* End of (? handling */        }       /* End of (? handling */
5855    
5856      /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,      /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
5857      all unadorned brackets become non-capturing and behave like (?:...)      is set, all unadorned brackets become non-capturing and behave like (?:...)
5858      brackets. */      brackets. */
5859    
5860      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)      else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
# Line 5242  we set the flag only if there is a liter Line 5872  we set the flag only if there is a liter
5872        skipbytes = 2;        skipbytes = 2;
5873        }        }
5874    
5875      /* Process nested bracketed regex. Assertions may not be repeated, but      /* Process nested bracketed regex. Assertions used not to be repeatable,
5876      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
5877      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
5878      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. */  
5879    
5880      previous = (bravalue >= OP_ONCE)? code : NULL;      previous = code;                   /* For handling repetition */
5881      *code = bravalue;      *code = bravalue;
5882      tempcode = code;      tempcode = code;
5883      tempreqvary = cd->req_varyopt;     /* Save value before bracket */      tempreqvary = cd->req_varyopt;     /* Save value before bracket */
# Line 5256  we set the flag only if there is a liter Line 5885  we set the flag only if there is a liter
5885    
5886      if (!compile_regex(      if (!compile_regex(
5887           newoptions,                   /* The complete new option state */           newoptions,                   /* The complete new option state */
          options & PCRE_IMS,           /* The previous ims option state */  
5888           &tempcode,                    /* Where to put code (updated) */           &tempcode,                    /* Where to put code (updated) */
5889           &ptr,                         /* Input pointer (updated) */           &ptr,                         /* Input pointer (updated) */
5890           errorcodeptr,                 /* Where to put an error message */           errorcodeptr,                 /* Where to put an error message */
# Line 5272  we set the flag only if there is a liter Line 5900  we set the flag only if there is a liter
5900             &length_prevgroup           /* Pre-compile phase */             &length_prevgroup           /* Pre-compile phase */
5901           ))           ))
5902        goto FAILED;        goto FAILED;
5903    
5904        if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT)
5905          cd->assert_depth -= 1;
5906    
5907      /* 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
5908      group, while tempcode has been updated to point past the end of the group      group, while tempcode has been updated to point past the end of the group
# Line 5343  we set the flag only if there is a liter Line 5974  we set the flag only if there is a liter
5974          goto FAILED;          goto FAILED;
5975          }          }
5976        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;        *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5977        *code++ = OP_BRA;        code++;   /* This already contains bravalue */
5978        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
5979        *code++ = OP_KET;        *code++ = OP_KET;
5980        PUTINC(code, 0, 1 + LINK_SIZE);        PUTINC(code, 0, 1 + LINK_SIZE);
# Line 5416  we set the flag only if there is a liter Line 6047  we set the flag only if there is a liter
6047    
6048      /* ===================================================================*/      /* ===================================================================*/
6049      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values      /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
6050      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
6051      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
6052      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
6053      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
6054      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
6055        ever created. */
6056    
6057      case CHAR_BACKSLASH:      case CHAR_BACKSLASH:
6058      tempptr = ptr;      tempptr = ptr;
# Line 5510  we set the flag only if there is a liter Line 6142  we set the flag only if there is a liter
6142          }          }
6143    
6144        /* \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).
6145        We also support \k{name} (.NET syntax) */        We also support \k{name} (.NET syntax).  */
6146    
6147        if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||        if (-c == ESC_k)
           ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))  
6148          {          {
6149            if ((ptr[1] != CHAR_LESS_THAN_SIGN &&
6150              ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET))
6151              {
6152              *errorcodeptr = ERR69;
6153              break;
6154              }
6155          is_recurse = FALSE;          is_recurse = FALSE;
6156          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?          terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
6157            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?            CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
6158            CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;            CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
6159          goto NAMED_REF_OR_RECURSE;          goto NAMED_REF_OR_RECURSE;
6160          }          }
6161    
6162        /* Back references are handled specially; must disable firstbyte if        /* Back references are handled specially; must disable firstbyte if
6163        not set to cope with cases like (?=(\w+))\1: which would otherwise set        not set to cope with cases like (?=(\w+))\1: which would otherwise set
# Line 5528  we set the flag only if there is a liter Line 6165  we set the flag only if there is a liter
6165    
6166        if (-c >= ESC_REF)        if (-c >= ESC_REF)
6167          {          {
6168            open_capitem *oc;
6169          recno = -c - ESC_REF;          recno = -c - ESC_REF;
6170    
6171          HANDLE_REFERENCE:    /* Come here from named backref handling */          HANDLE_REFERENCE:    /* Come here from named backref handling */
6172          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
6173          previous = code;          previous = code;
6174          *code++ = OP_REF;          *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF;
6175          PUT2INC(code, 0, recno);          PUT2INC(code, 0, recno);
6176          cd->backref_map |= (recno < 32)? (1 << recno) : 1;          cd->backref_map |= (recno < 32)? (1 << recno) : 1;
6177          if (recno > cd->top_backref) cd->top_backref = recno;          if (recno > cd->top_backref) cd->top_backref = recno;
6178    
6179            /* Check to see if this back reference is recursive, that it, it
6180            is inside the group that it references. A flag is set so that the
6181            group can be made atomic. */
6182    
6183            for (oc = cd->open_caps; oc != NULL; oc = oc->next)
6184              {
6185              if (oc->number == recno)
6186                {
6187                oc->flag = TRUE;
6188                break;
6189                }
6190              }
6191          }          }
6192    
6193        /* So are Unicode property matches, if supported. */        /* So are Unicode property matches, if supported. */
# Line 5566  we set the flag only if there is a liter Line 6217  we set the flag only if there is a liter
6217  #endif  #endif
6218    
6219        /* For the rest (including \X when Unicode properties are supported), we        /* For the rest (including \X when Unicode properties are supported), we
6220        can obtain the OP value by negating the escape value. */        can obtain the OP value by negating the escape value in the default
6221          situation when PCRE_UCP is not set. When it *is* set, we substitute
6222          Unicode property tests. */
6223    
6224        else        else
6225          {          {
6226          previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;  #ifdef SUPPORT_UCP
6227          *code++ = -c;          if (-c >= ESC_DU && -c <= ESC_wu)
6228              {
6229              nestptr = ptr + 1;                   /* Where to resume */
6230              ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
6231              }
6232            else
6233    #endif
6234              {
6235              previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
6236              *code++ = -c;
6237              }
6238          }          }
6239        continue;        continue;
6240        }        }
# Line 5616  we set the flag only if there is a liter Line 6279  we set the flag only if there is a liter
6279    
6280      ONE_CHAR:      ONE_CHAR:
6281      previous = code;      previous = code;
6282      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;      *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR;
6283      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];      for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6284    
6285      /* Remember if \r or \n were seen */      /* Remember if \r or \n were seen */
# Line 5680  return FALSE; Line 6343  return FALSE;
6343  /* 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
6344  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
6345  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.  
   
6346  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
6347  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
6348  value of lengthptr distinguishes the two phases.  value of lengthptr distinguishes the two phases.
6349    
6350  Arguments:  Arguments:
6351    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  
6352    codeptr        -> the address of the current code pointer    codeptr        -> the address of the current code pointer
6353    ptrptr         -> the address of the current pattern pointer    ptrptr         -> the address of the current pattern pointer
6354    errorcodeptr   -> pointer to error code variable    errorcodeptr   -> pointer to error code variable
# Line 5709  Returns: TRUE on success Line 6366  Returns: TRUE on success
6366  */  */
6367    
6368  static BOOL  static BOOL
6369  compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,  compile_regex(int options, uschar **codeptr, const uschar **ptrptr,
6370    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,    int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6371    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,    int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6372    int *lengthptr)    int *lengthptr)
# Line 5729  int max_bracount; Line 6386  int max_bracount;
6386  branch_chain bc;  branch_chain bc;
6387    
6388  bc.outer = bcptr;  bc.outer = bcptr;
6389  bc.current = code;  bc.current_branch = code;
6390    
6391  firstbyte = reqbyte = REQ_UNSET;  firstbyte = reqbyte = REQ_UNSET;
6392    
# Line 5748  them global. It tests the value of lengt Line 6405  them global. It tests the value of lengt
6405  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. */
6406    
6407  /* 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
6408  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
6409    detect groups that contain recursive back references to themselves. Note that
6410    only OP_CBRA need be tested here; changing this opcode to one of its variants,
6411    e.g. OP_SCBRAPOS, happens later, after the group has been compiled. */
6412    
6413  if (*code == OP_CBRA)  if (*code == OP_CBRA)
6414    {    {
6415    capnumber = GET2(code, 1 + LINK_SIZE);    capnumber = GET2(code, 1 + LINK_SIZE);
6416    capitem.number = capnumber;    capitem.number = capnumber;
6417    capitem.next = cd->open_caps;    capitem.next = cd->open_caps;
6418    cd->open_caps = &capitem;    capitem.flag = FALSE;
6419    }    cd->open_caps = &capitem;
6420      }
6421    
6422  /* Offset is set zero to mark that this bracket is still open */  /* Offset is set zero to mark that this bracket is still open */
6423    
# Line 5773  for (;;) Line 6434  for (;;)
6434    
6435    if (reset_bracount) cd->bracount = orig_bracount;    if (reset_bracount) cd->bracount = orig_bracount;
6436    
   /* 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;  
     }  
   
6437    /* Set up dummy OP_REVERSE if lookbehind assertion */    /* Set up dummy OP_REVERSE if lookbehind assertion */
6438    
6439    if (lookbehind)    if (lookbehind)
# Line 5852  for (;;) Line 6504  for (;;)
6504    
6505      /* If lookbehind, check that this branch matches a fixed-length string, and      /* If lookbehind, check that this branch matches a fixed-length string, and
6506      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
6507      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
6508      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
6509      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
6510      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
6511      the problem can be shown. */      the problem can be shown. */
6512    
6513      if (lookbehind)      if (lookbehind)
6514        {        {
6515        int fixed_length;        int fixed_length;
6516        *code = OP_END;        *code = OP_END;
6517        fixed_length = find_fixedlength(last_branch, options, FALSE, cd);        fixed_length = find_fixedlength(last_branch,  (options & PCRE_UTF8) != 0,
6518            FALSE, cd);
6519        DPRINTF(("fixed length = %d\n", fixed_length));        DPRINTF(("fixed length = %d\n", fixed_length));
6520        if (fixed_length == -3)        if (fixed_length == -3)
6521          {          {
6522          cd->check_lookbehind = TRUE;          cd->check_lookbehind = TRUE;
6523          }          }
6524        else if (fixed_length < 0)        else if (fixed_length < 0)
6525          {          {
6526          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;          *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
# Line 5883  for (;;) Line 6536  for (;;)
6536    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
6537    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
6538    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
6539    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. */  
6540    
6541    if (*ptr != CHAR_VERTICAL_LINE)    if (*ptr != CHAR_VERTICAL_LINE)
6542      {      {
6543      if (lengthptr == NULL)      if (lengthptr == NULL)
6544        {        {
6545        int branch_length = code - last_branch;        int branch_length = (int)(code - last_branch);
6546        do        do
6547          {          {
6548          int prev_length = GET(last_branch, 1);          int prev_length = GET(last_branch, 1);
# Line 5901  for (;;) Line 6552  for (;;)
6552          }          }
6553        while (branch_length > 0);        while (branch_length > 0);
6554        }        }
   
     /* If it was a capturing subpattern, remove it from the chain. */  
   
     if (capnumber > 0) cd->open_caps = cd->open_caps->next;  
6555    
6556      /* Fill in the ket */      /* Fill in the ket */
6557    
6558      *code = OP_KET;      *code = OP_KET;
6559      PUT(code, 1, code - start_bracket);      PUT(code, 1, (int)(code - start_bracket));
6560      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6561    
6562      /* Resetting option if needed */      /* If it was a capturing subpattern, check to see if it contained any
6563        recursive back references. If so, we must wrap it in atomic brackets.
6564        In any event, remove the block from the chain. */
6565    
6566      if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)      if (capnumber > 0)
6567        {        {
6568        *code++ = OP_OPT;        if (cd->open_caps->flag)
6569        *code++ = oldims;          {
6570        length += 2;          memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6571              code - start_bracket);
6572            *start_bracket = OP_ONCE;
6573            code += 1 + LINK_SIZE;
6574            PUT(start_bracket, 1, (int)(code - start_bracket));
6575            *code = OP_KET;
6576            PUT(code, 1, (int)(code - start_bracket));
6577            code += 1 + LINK_SIZE;
6578            length += 2 + 2*LINK_SIZE;
6579            }
6580          cd->open_caps = cd->open_caps->next;
6581        }        }
6582    
6583      /* Retain the highest bracket number, in case resetting was used. */      /* Retain the highest bracket number, in case resetting was used. */
# Line 5960  for (;;) Line 6619  for (;;)
6619    else    else
6620      {      {
6621      *code = OP_ALT;      *code = OP_ALT;
6622      PUT(code, 1, code - last_branch);      PUT(code, 1, (int)(code - last_branch));
6623      bc.current = last_branch = code;      bc.current_branch = last_branch = code;
6624      code += 1 + LINK_SIZE;      code += 1 + LINK_SIZE;
6625      }      }
6626    
# Line 5980  for (;;) Line 6639  for (;;)
6639  /* 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
6640  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
6641  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
6642  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
6643  counts, since OP_CIRC can match in the middle.  be found, because ^ generates OP_CIRCM in that mode.
6644    
6645  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.
6646  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 6002  of the more common cases more precisely. Line 6661  of the more common cases more precisely.
6661    
6662  Arguments:  Arguments:
6663    code           points to start of expression (the bracket)    code           points to start of expression (the bracket)
   options        points to the options setting  
6664    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
6665                    handles up to substring 31; after that we just have to take                    handles up to substring 31; after that we just have to take
6666                    the less precise approach                    the less precise approach
# Line 6012  Returns: TRUE or FALSE Line 6670  Returns: TRUE or FALSE
6670  */  */
6671    
6672  static BOOL  static BOOL
6673  is_anchored(register const uschar *code, int *options, unsigned int bracket_map,  is_anchored(register const uschar *code, unsigned int bracket_map,
6674    unsigned int backref_map)    unsigned int backref_map)
6675  {  {
6676  do {  do {
6677     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6678       options, PCRE_MULTILINE, FALSE);       FALSE);
6679     register int op = *scode;     register int op = *scode;
6680    
6681     /* Non-capturing brackets */     /* Non-capturing brackets */
6682    
6683     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
6684           op == OP_SBRA || op == OP_SBRAPOS)
6685       {       {
6686       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6687       }       }
6688    
6689     /* Capturing brackets */     /* Capturing brackets */
6690    
6691     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
6692                op == OP_SCBRA || op == OP_SCBRAPOS)
6693       {       {
6694       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
6695       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6696       if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;       if (!is_anchored(scode, new_map, backref_map)) return FALSE;
6697       }       }
6698    
6699     /* Other brackets */     /* Other brackets */
6700    
6701     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)     else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6702       {       {
6703       if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;       if (!is_anchored(scode, bracket_map, backref_map)) return FALSE;
6704       }       }
6705    
6706     /* .* 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 6055  do { Line 6715  do {
6715    
6716     /* Check for explicit anchoring */     /* Check for explicit anchoring */
6717    
6718     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;  
6719     code += GET(code, 1);     code += GET(code, 1);
6720     }     }
6721  while (*code == OP_ALT);   /* Loop for each alternative */  while (*code == OP_ALT);   /* Loop for each alternative */
# Line 6093  is_startline(const uschar *code, unsigne Line 6751  is_startline(const uschar *code, unsigne
6751  {  {
6752  do {  do {
6753     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],     const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6754       NULL, 0, FALSE);       FALSE);
6755     register int op = *scode;     register int op = *scode;
6756    
6757     /* 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 6108  do { Line 6766  do {
6766       switch (*scode)       switch (*scode)
6767         {         {
6768         case OP_CREF:         case OP_CREF:
6769           case OP_NCREF:
6770         case OP_RREF:         case OP_RREF:
6771           case OP_NRREF:
6772         case OP_DEF:         case OP_DEF:
6773         return FALSE;         return FALSE;
6774    
# Line 6118  do { Line 6778  do {
6778         scode += 1 + LINK_SIZE;         scode += 1 + LINK_SIZE;
6779         break;         break;
6780         }         }
6781       scode = first_significant_code(scode, NULL, 0, FALSE);       scode = first_significant_code(scode, FALSE);
6782       op = *scode;       op = *scode;
6783       }       }
6784    
6785     /* Non-capturing brackets */     /* Non-capturing brackets */
6786    
6787     if (op == OP_BRA)     if (op == OP_BRA  || op == OP_BRAPOS ||
6788           op == OP_SBRA || op == OP_SBRAPOS)
6789       {       {
6790       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;       if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6791       }       }
6792    
6793     /* Capturing brackets */     /* Capturing brackets */
6794    
6795     else if (op == OP_CBRA)     else if (op == OP_CBRA  || op == OP_CBRAPOS ||
6796                op == OP_SCBRA || op == OP_SCBRAPOS)
6797       {       {
6798       int n = GET2(scode, 1+LINK_SIZE);       int n = GET2(scode, 1+LINK_SIZE);
6799       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);       int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
# Line 6155  do { Line 6817  do {
6817    
6818     /* Check for explicit circumflex */     /* Check for explicit circumflex */
6819    
6820     else if (op != OP_CIRC) return FALSE;     else if (op != OP_CIRC && op != OP_CIRCM) return FALSE;
6821    
6822     /* Move on to the next alternative */     /* Move on to the next alternative */
6823    
# Line 6181  we return that char, otherwise -1. Line 6843  we return that char, otherwise -1.
6843    
6844  Arguments:  Arguments:
6845    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)  
6846    inassert   TRUE if in an assertion    inassert   TRUE if in an assertion
6847    
6848  Returns:     -1 or the fixed first char  Returns:     -1 or the fixed first char
6849  */  */
6850    
6851  static int  static int
6852  find_firstassertedchar(const uschar *code, int *options, BOOL inassert)  find_firstassertedchar(const uschar *code, BOOL inassert)
6853  {  {
6854  register int c = -1;  register int c = -1;
6855  do {  do {
6856     int d;     int d;
6857     const uschar *scode =     int xl = (*code == OP_CBRA || *code == OP_SCBRA ||
6858       first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);               *code == OP_CBRAPOS || *code == OP_SCBRAPOS)? 2:0;
6859       const uschar *scode = first_significant_code(code + 1+LINK_SIZE + xl, TRUE);
6860     register int op = *scode;     register int op = *scode;
6861    
6862     switch(op)     switch(op)
# Line 6203  do { Line 6865  do {
6865       return -1;       return -1;
6866    
6867       case OP_BRA:       case OP_BRA:
6868         case OP_BRAPOS:
6869       case OP_CBRA:       case OP_CBRA:
6870         case OP_SCBRA:
6871         case OP_CBRAPOS:
6872         case OP_SCBRAPOS:
6873       case OP_ASSERT:       case OP_ASSERT:
6874       case OP_ONCE:       case OP_ONCE:
6875       case OP_COND:       case OP_COND:
6876       if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)       if ((d = find_firstassertedchar(scode, op == OP_ASSERT)) < 0)
6877         return -1;         return -1;
6878       if (c < 0) c = d; else if (c != d) return -1;       if (c < 0) c = d; else if (c != d) return -1;
6879       break;       break;
6880    
6881       case OP_EXACT:       /* Fall through */       case OP_EXACT:
6882       scode += 2;       scode += 2;
6883         /* Fall through */
6884    
6885       case OP_CHAR:       case OP_CHAR:
      case OP_CHARNC:  
6886       case OP_PLUS:       case OP_PLUS:
6887       case OP_MINPLUS:       case OP_MINPLUS:
6888       case OP_POSPLUS:       case OP_POSPLUS:
6889       if (!inassert) return -1;       if (!inassert) return -1;
6890       if (c < 0)       if (c < 0) c = scode[1];
6891         {         else if (c != scode[1]) return -1;
6892         c = scode[1];       break;
6893         if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;  
6894         }       case OP_EXACTI:
6895       else if (c != scode[1]) return -1;       scode += 2;
6896         /* Fall through */
6897    
6898         case OP_CHARI:
6899         case OP_PLUSI:
6900         case OP_MINPLUSI:
6901         case OP_POSPLUSI:
6902         if (!inassert) return -1;
6903         if (c < 0) c = scode[1] | REQ_CASELESS;
6904           else if (c != scode[1]) return -1;
6905       break;       break;
6906       }       }
6907    
# Line 6277  int length = 1; /* For final END opcode Line 6952  int length = 1; /* For final END opcode
6952  int firstbyte, reqbyte, newline;  int firstbyte, reqbyte, newline;
6953  int errorcode = 0;  int errorcode = 0;
6954  int skipatstart = 0;  int skipatstart = 0;
6955  BOOL utf8 = (options & PCRE_UTF8) != 0;  BOOL utf8;
6956  size_t size;  size_t size;
6957  uschar *code;  uschar *code;
6958  const uschar *codestart;  const uschar *codestart;
# Line 6347  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7022  while (ptr[skipatstart] == CHAR_LEFT_PAR
7022    
7023    if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)    if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
7024      { skipatstart += 7; options |= PCRE_UTF8; continue; }      { skipatstart += 7; options |= PCRE_UTF8; continue; }
7025      else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
7026        { skipatstart += 6; options |= PCRE_UCP; continue; }
7027      else if (strncmp((char *)(ptr+skipatstart+2), STRING_NO_START_OPT_RIGHTPAR, 13) == 0)
7028        { skipatstart += 15; options |= PCRE_NO_START_OPTIMIZE; continue; }
7029    
7030    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)    if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
7031      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }      { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
# Line 6371  while (ptr[skipatstart] == CHAR_LEFT_PAR Line 7050  while (ptr[skipatstart] == CHAR_LEFT_PAR
7050    else break;    else break;
7051    }    }
7052    
7053  /* Can't support UTF8 unless PCRE has been compiled to include the code. */  utf8 = (options & PCRE_UTF8) != 0;
7054    
7055    /* Can't support UTF8 unless PCRE has been compiled to include the code. The
7056    return of an error code from _pcre_valid_utf8() is a new feature, introduced in
7057    release 8.13. It is passed back from pcre_[dfa_]exec(), but at the moment is
7058    not used here. */
7059    
7060  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
7061  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&  if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
7062       (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)       (errorcode = _pcre_valid_utf8((USPTR)pattern, -1, erroroffset)) != 0)
7063    {    {
7064    errorcode = ERR44;    errorcode = ERR44;
7065    goto PCRE_EARLY_ERROR_RETURN2;    goto PCRE_EARLY_ERROR_RETURN2;
# Line 6388  if (utf8) Line 7072  if (utf8)
7072    }    }
7073  #endif  #endif
7074    
7075    /* Can't support UCP unless PCRE has been compiled to include the code. */
7076    
7077    #ifndef SUPPORT_UCP
7078    if ((options & PCRE_UCP) != 0)
7079      {
7080      errorcode = ERR67;
7081      goto PCRE_EARLY_ERROR_RETURN;
7082      }
7083    #endif
7084    
7085  /* Check validity of \R options. */  /* Check validity of \R options. */
7086    
7087  switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))  switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
# Line 6481  outside can help speed up starting point Line 7175  outside can help speed up starting point
7175  ptr += skipatstart;  ptr += skipatstart;
7176  code = cworkspace;  code = cworkspace;
7177  *code = OP_BRA;  *code = OP_BRA;
7178  (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,  (void)compile_regex(cd->external_options, &code, &ptr, &errorcode, FALSE,
7179    &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,    FALSE, 0, &firstbyte, &reqbyte, NULL, cd, &length);
   &length);  
7180  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;  if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
7181    
7182  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,  DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
# Line 6516  regex compiled on a system with 4-byte p Line 7209  regex compiled on a system with 4-byte p
7209  pointers. */  pointers. */
7210    
7211  re->magic_number = MAGIC_NUMBER;  re->magic_number = MAGIC_NUMBER;
7212  re->size = size;  re->size = (int)size;
7213  re->options = cd->external_options;  re->options = cd->external_options;
7214  re->flags = cd->external_flags;  re->flags = cd->external_flags;
7215  re->dummy1 = 0;  re->dummy1 = 0;
# Line 6537  field; this time it's used for rememberi Line 7230  field; this time it's used for rememberi
7230  */  */
7231    
7232  cd->final_bracount = cd->bracount;  /* Save for checking forward references */  cd->final_bracount = cd->bracount;  /* Save for checking forward references */
7233    cd->assert_depth = 0;
7234  cd->bracount = 0;  cd->bracount = 0;
7235  cd->names_found = 0;  cd->names_found = 0;
7236  cd->name_table = (uschar *)re + re->name_table_offset;  cd->name_table = (uschar *)re + re->name_table_offset;
# Line 6555  of the function here. */ Line 7249  of the function here. */
7249  ptr = (const uschar *)pattern + skipatstart;  ptr = (const uschar *)pattern + skipatstart;
7250  code = (uschar *)codestart;  code = (uschar *)codestart;
7251  *code = OP_BRA;  *code = OP_BRA;
7252  (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,  (void)compile_regex(re->options, &code, &ptr, &errorcode, FALSE, FALSE, 0,
7253    &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);    &firstbyte, &reqbyte, NULL, cd, NULL);
7254  re->top_bracket = cd->bracount;  re->top_bracket = cd->bracount;
7255  re->top_backref = cd->top_backref;  re->top_backref = cd->top_backref;
7256  re->flags = cd->external_flags;  re->flags = cd->external_flags;
# Line 6572  if debugging, leave the test till after Line 7266  if debugging, leave the test till after
7266    
7267  *code++ = OP_END;  *code++ = OP_END;
7268