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

Diff of /code/trunk/pcre_compile.c

Parent Directory | Revision Log | View Patch Patch

-revision 298 by ph10,
Thu Jan 10 17:09:12 2008 UTC
+revision 336 by ph10,
Sat Apr 12 15:59:03 2008 UTC
 Line 6
  and semantics are as close as possible to those of the Perl 5 language.
                         Written by Philip Hazel
-            Copyright (c) 1997-2007 University of Cambridge
+            Copyright (c) 1997-2008 University of Cambridge
  -----------------------------------------------------------------------------
  Redistribution and use in source and binary forms, with or without
 Line 158 
 static const char verbnames[] =
    "SKIP\0"
    "THEN";
- static verbitem verbs[] = {
+ static const verbitem verbs[] = {
    { 6, OP_ACCEPT },
    { 6, OP_COMMIT },
    { 1, OP_FAIL },
 Line 168 
 static verbitem verbs[] = {
    { 4, OP_THEN  }
  };
- static int verbcount = sizeof(verbs)/sizeof(verbitem);
+ static const int verbcount = sizeof(verbs)/sizeof(verbitem);
  /* Tables of names of POSIX character classes and their lengths. The names are
 Line 295 
 static const char error_texts[] =
    /* 55 */
    "repeating a DEFINE group is not allowed\0"
    "inconsistent NEWLINE options\0"
-   "\\g is not followed by a braced name or an optionally braced non-zero number\0"
+   "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
-   "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number\0"
+   "a numbered reference must not be zero\0"
    "(*VERB) with an argument is not supported\0"
    /* 60 */
    "(*VERB) not recognized\0"
    "number is too big\0"
    "subpattern name expected\0"
-   "digit expected after (?+";
+   "digit expected after (?+\0"
+   "] is an invalid data character in JavaScript compatibility mode";
  /* Table to identify digits and hex digits. This is used when compiling
-Line 531 
 else
+Line 532 
 else
      *errorcodeptr = ERR37;
      break;
-     /* \g must be followed by a number, either plain or braced. If positive, it
+     /* \g must be followed by one of a number of specific things:
-     is an absolute backreference. If negative, it is a relative backreference.
-     This is a Perl 5.10 feature. Perl 5.10 also supports \g{name} as a
+     (1) A number, either plain or braced. If positive, it is an absolute
-     reference to a named group. This is part of Perl's movement towards a
+     backreference. If negative, it is a relative backreference. This is a Perl
-     unified syntax for back references. As this is synonymous with \k{name}, we
+.10 feature.
-     fudge it up by pretending it really was \k. */
+     (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
+     is part of Perl's movement towards a unified syntax for back references. As
+     this is synonymous with \k{name}, we fudge it up by pretending it really
+     was \k.
+     (3) For Oniguruma compatibility we also support \g followed by a name or a
+     number either in angle brackets or in single quotes. However, these are
+     (possibly recursive) subroutine calls, _not_ backreferences. Just return
+     the -ESC_g code (cf \k). */
      case 'g':
+     if (ptr[1] == '<' || ptr[1] == '\'')
+       {
+       c = -ESC_g;
+       break;
+       }
+     /* Handle the Perl-compatible cases */
      if (ptr[1] == '{')
        {
        const uschar *p;
-Line 565 
 else
+Line 583 
 else
      while ((digitab[ptr[1]] & ctype_digit) != 0)
        c = c * 10 + *(++ptr) - '0';
-     if (c < 0)
+     if (c < 0)   /* Integer overflow */
        {
        *errorcodeptr = ERR61;
        break;
        }
-     if (c == 0 || (braced && *(++ptr) != '}'))
+     if (braced && *(++ptr) != '}')
        {
        *errorcodeptr = ERR57;
        break;
        }
+     if (c == 0)
+       {
+       *errorcodeptr = ERR58;
+       break;
+       }
      if (negated)
        {
-Line 611 
 else
+Line 635 
 else
        c -= '0';
        while ((digitab[ptr[1]] & ctype_digit) != 0)
          c = c * 10 + *(++ptr) - '0';
-       if (c < 0)
+       if (c < 0)    /* Integer overflow */
          {
          *errorcodeptr = ERR61;
          break;
-Line 1544 
 for (code = first_significant_code(code
+Line 1568 
 for (code = first_significant_code(code
    /* Groups with zero repeats can of course be empty; skip them. */
-   if (c == OP_BRAZERO || c == OP_BRAMINZERO)
+   if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
      {
      code += _pcre_OP_lengths[c];
      do code += GET(code, 1); while (*code == OP_ALT);
-Line 1824 
 return -1;
+Line 1848 
 return -1;
  that is referenced. This means that groups can be replicated for fixed
  repetition simply by copying (because the recursion is allowed to refer to
  earlier groups that are outside the current group). However, when a group is
- optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
+ optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
- it, after it has been compiled. This means that any OP_RECURSE items within it
+ inserted before it, after it has been compiled. This means that any OP_RECURSE
- that refer to the group itself or any contained groups have to have their
+ items within it that refer to the group itself or any contained groups have to
- offsets adjusted. That one of the jobs of this function. Before it is called,
+ have their offsets adjusted. That one of the jobs of this function. Before it
- the partially compiled regex must be temporarily terminated with OP_END.
+ is called, the partially compiled regex must be temporarily terminated with
+ OP_END.
  This function has been extended with the possibility of forward references for
  recursions and subroutine calls. It must also check the list of such references
-Line 1859 
 while ((ptr = (uschar *)find_recurse(ptr
+Line 1884 
 while ((ptr = (uschar *)find_recurse(ptr
    /* See if this recursion is on the forward reference list. If so, adjust the
    reference. */
    for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
      {
      offset = GET(hc, 0);
-Line 2113 
 if (next >= 0) switch(op_code)
+Line 2138 
 if (next >= 0) switch(op_code)
    /* For OP_NOT, "item" must be a single-byte character. */
    case OP_NOT:
-   if (next < 0) return FALSE;  /* Not a character */
    if (item == next) return TRUE;
    if ((options & PCRE_CASELESS) == 0) return FALSE;
  #ifdef SUPPORT_UTF8
-Line 2376 
 uschar classbits[32];
+Line 2400 
 uschar classbits[32];
  BOOL class_utf8;
  BOOL utf8 = (options & PCRE_UTF8) != 0;
  uschar *class_utf8data;
+ uschar *class_utf8data_base;
  uschar utf8_char[6];
  #else
  BOOL utf8 = FALSE;
-Line 2435 
 for (;; ptr++)
+Line 2460 
 for (;; ptr++)
    /* Get next byte in the pattern */
    c = *ptr;
    /* If we are in the pre-compile phase, accumulate the length used for the
    previous cycle of this loop. */
-Line 2630 
 for (;; ptr++)
+Line 2655 
 for (;; ptr++)
      opcode is compiled. It may optionally have a bit map for characters < 256,
      but those above are are explicitly listed afterwards. A flag byte tells
      whether the bitmap is present, and whether this is a negated class or not.
-     */
+     In JavaScript compatibility mode, an isolated ']' causes an error. In
+     default (Perl) mode, it is treated as a data character. */
+     case ']':
+     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
+       {
+       *errorcodeptr = ERR64;
+       goto FAILED;
+       }
+     goto NORMAL_CHAR;
      case '[':
      previous = code;
-Line 2687 
 for (;; ptr++)
+Line 2722 
 for (;; ptr++)
  #ifdef SUPPORT_UTF8
      class_utf8 = FALSE;                       /* No chars >= 256 */
      class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
+     class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
  #endif
      /* Process characters until ] is reached. By writing this as a "do" it
-Line 2702 
 for (;; ptr++)
+Line 2738 
 for (;; ptr++)
          {                           /* Braces are required because the */
          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
          }
+       /* In the pre-compile phase, accumulate the length of any UTF-8 extra
+       data and reset the pointer. This is so that very large classes that
+       contain a zillion UTF-8 characters no longer overwrite the work space
+       (which is on the stack). */
+       if (lengthptr != NULL)
+         {
+         *lengthptr += class_utf8data - class_utf8data_base;
+         class_utf8data = class_utf8data_base;
+         }
  #endif
        /* Inside \Q...\E everything is literal except \E */
-Line 3806 
 we set the flag only if there is a liter
+Line 3854 
 we set the flag only if there is a liter
        if (repeat_min == 0)
          {
-         /* If the maximum is also zero, we just omit the group from the output
+         /* If the maximum is also zero, we used to just omit the group from the
-         altogether. */
+         output altogether, like this:
-         if (repeat_max == 0)
+         ** if (repeat_max == 0)
-           {
+         **   {
-           code = previous;
+         **   code = previous;
-           goto END_REPEAT;
+         **   goto END_REPEAT;
-           }
+         **   }
+         However, that fails when a group is referenced as a subroutine from
+         elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
+         so that it is skipped on execution. As we don't have a list of which
+         groups are referenced, we cannot do this selectively.
+         If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
+         and do no more at this point. However, we do need to adjust any
+         OP_RECURSE calls inside the group that refer to the group itself or any
+         internal or forward referenced group, because the offset is from the
+         start of the whole regex. Temporarily terminate the pattern while doing
+         this. */
-         /* If the maximum is 1 or unlimited, we just have to stick in the
+         if (repeat_max <= 1)    /* Covers 0, 1, and unlimited */
-         BRAZERO and do no more at this point. However, we do need to adjust
-         any OP_RECURSE calls inside the group that refer to the group itself or
-         any internal or forward referenced group, because the offset is from
-         the start of the whole regex. Temporarily terminate the pattern while
-         doing this. */
-         if (repeat_max <= 1)
            {
            *code = OP_END;
            adjust_recurse(previous, 1, utf8, cd, save_hwm);
            memmove(previous+1, previous, len);
            code++;
+           if (repeat_max == 0)
+             {
+             *previous++ = OP_SKIPZERO;
+             goto END_REPEAT;
+             }
            *previous++ = OP_BRAZERO + repeat_type;
            }
-Line 4102 
 we set the flag only if there is a liter
+Line 4160 
 we set the flag only if there is a liter
      bravalue = OP_CBRA;
      save_hwm = cd->hwm;
      reset_bracount = FALSE;
      /* First deal with various "verbs" that can be introduced by '*'. */
      if (*(++ptr) == '*' && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
-Line 4554 
 we set the flag only if there is a liter
+Line 4612 
 we set the flag only if there is a liter
          references (?P=name) and recursion (?P>name), as well as falling
          through from the Perl recursion syntax (?&name). We also come here from
          the Perl \k<name> or \k'name' back reference syntax and the \k{name}
-         .NET syntax. */
+         .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
          NAMED_REF_OR_RECURSE:
          name = ++ptr;
-Line 4632 
 we set the flag only if there is a liter
+Line 4690 
 we set the flag only if there is a liter
          case '5': case '6': case '7': case '8': case '9':   /* subroutine */
            {
            const uschar *called;
+           terminator = ')';
+           /* Come here from the \g<...> and \g'...' code (Oniguruma
+           compatibility). However, the syntax has been checked to ensure that
+           the ... are a (signed) number, so that neither ERR63 nor ERR29 will
+           be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
+           ever be taken. */
+           HANDLE_NUMERICAL_RECURSION:
            if ((refsign = *ptr) == '+')
              {
-Line 4653 
 we set the flag only if there is a liter
+Line 4720 
 we set the flag only if there is a liter
            while((digitab[*ptr] & ctype_digit) != 0)
              recno = recno * 10 + *ptr++ - '0';
-           if (*ptr != ')')
+           if (*ptr != terminator)
              {
              *errorcodeptr = ERR29;
              goto FAILED;
-Line 5049 
 we set the flag only if there is a liter
+Line 5116 
 we set the flag only if there is a liter
      back references and those types that consume a character may be repeated.
      We can test for values between ESC_b and ESC_Z for the latter; this may
      have to change if any new ones are ever created. */
      case '\\':
      tempptr = ptr;
      c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
-Line 5076 
 we set the flag only if there is a liter
+Line 5143 
 we set the flag only if there is a liter
        zerofirstbyte = firstbyte;
        zeroreqbyte = reqbyte;
+       /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
+       is a subroutine call by number (Oniguruma syntax). In fact, the value
+       -ESC_g is returned only for these cases. So we don't need to check for <
+       or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
+       -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
+       that is a synonym for a named back reference). */
+       if (-c == ESC_g)
+         {
+         const uschar *p;
+         save_hwm = cd->hwm;   /* Normally this is set when '(' is read */
+         terminator = (*(++ptr) == '<')? '>' : '\'';
+         /* These two statements stop the compiler for warning about possibly
+         unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
+         fact, because we actually check for a number below, the paths that
+         would actually be in error are never taken. */
+         skipbytes = 0;
+         reset_bracount = FALSE;
+         /* Test for a name */
+         if (ptr[1] != '+' && ptr[1] != '-')
+           {
+           BOOL isnumber = TRUE;
+           for (p = ptr + 1; *p != 0 && *p != terminator; p++)
+             {
+             if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
+             if ((cd->ctypes[*p] & ctype_word) == 0) break;
+             }
+           if (*p != terminator)
+             {
+             *errorcodeptr = ERR57;
+             break;
+             }
+           if (isnumber)
+             {
+             ptr++;
+             goto HANDLE_NUMERICAL_RECURSION;
+             }
+           is_recurse = TRUE;
+           goto NAMED_REF_OR_RECURSE;
+           }
+         /* Test a signed number in angle brackets or quotes. */
+         p = ptr + 2;
+         while ((digitab[*p] & ctype_digit) != 0) p++;
+         if (*p != terminator)
+           {
+           *errorcodeptr = ERR57;
+           break;
+           }
+         ptr++;
+         goto HANDLE_NUMERICAL_RECURSION;
+         }
        /* \k<name> or \k'name' is a back reference by name (Perl syntax).
        We also support \k{name} (.NET syntax) */
-Line 5807 
 to fill in forward references to subpatt
+Line 5932 
 to fill in forward references to subpatt
  uschar cworkspace[COMPILE_WORK_SIZE];
  /* Set this early so that early errors get offset 0. */
  ptr = (const uschar *)pattern;
-Line 6096 
 while (errorcode == 0 && cd->hwm > cwork
+Line 6220 
 while (errorcode == 0 && cd->hwm > cwork
    if (groupptr == NULL) errorcode = ERR53;
      else PUT(((uschar *)codestart), offset, groupptr - codestart);
    }
  /* Give an error if there's back reference to a non-existent capturing
  subpattern. */

 Legend:



Removed from v.298
 


changed lines


 
Added in v.336
 Legend:



Removed from v.298
 


changed lines


 
Added in v.336
-Removed from v.298
+Added in v.336

webmaster@exim.org	ViewVC Help
Powered by ViewVC 1.1.12