[PATCH] Fix fnmatch escape handling in brackets (BZ #361)

[Markus F.X.J. Oberhumer]

I happily accept that FNM_CASEFOLD ranges are somewhat limited, but the docs 
need some clarificataion then; especially that GNU sed/grep/egrep and Perl 
_do_ get it right.

Please see the somewhat polished example program attached below.

Unrelated, I may have hit a bug in GNU sed 4.1.1 (works with 3.0.2 and 4.0.5). 
Please have a look as well.

Finally one could also argue that the following pseudocode would match the 
semantics better:

   if (ch >= range_start && char <= range_end) goto matched_range;
   if (flag_FNM_CASEFOLD)
      if (FOLD(ch) >= FOLD(range_start) && 
         FOLD(ch) <= FOLD(range_end)) goto matched_range;
   /* range not matched */

Markus

On Thursday 02 September 2004 10:22, Jakub Jelinek wrote:
> On Thu, Sep 02, 2004 at 05:51:18AM +0200, Markus F.X.J. Oberhumer wrote:
> > Jakub,
> >
> > I have not been able to test your patch yet, but the program attached
> > below prints some errors for me (I originally thought this was caused by
> > the backslash at the end but I now see it is completely unrelated).
> >
> > Not sure if this is actually a bug - at least it is confusing that when
> > enlarging the pattern-range or adding FNM_CASEFOLD an "A" does
> > not match a range starting with "A" anymore.
>
> I don't think it is a bug.
> Unlike RE_ICASE regcomp which converts to uppercase, FNM_CASEFOLD
> converts to lowercase.
> And a range where start character is > end character is invalid.
> You can play with regex RE_ICASE too:
>
> $ echo | LC_ALL=C sed -n '/[a-[]/Ip'
> $ echo | LC_ALL=C sed -n '/[A-[]/Ip'
> $ echo | LC_ALL=C sed -n '/[[-a]/Ip'
> sed: -e expression #1, char 8: Invalid range end
> $ echo | LC_ALL=C sed -n '/[[-A]/Ip'
> sed: -e expression #1, char 8: Invalid range end
> $ echo | LC_ALL=C sed -n '/[[-A]/p'
> sed: -e expression #1, char 7: Invalid range end
> $ echo | LC_ALL=C sed -n '/[[-a]/p'
> $ echo | LC_ALL=C sed -n '/[a-[]/p'
> sed: -e expression #1, char 7: Invalid range end
> $ echo | LC_ALL=C sed -n '/[A-[]/p'
>
> and fnmatch behaves very similarly to this (just with the difference
> that it uses lowercase instead of uppercase conversion).
> See that while [[-a] is valid range without RE_ICASE, it is not
> with RE_ICASE (but [a-[] is, which without RE_ICASE is not valid).
>
> So, when you are using FNM_CASEFOLD, it is always better to write the
> ranges where both range ends aren't a-z or A-Z in lowercase.
>
> I have checked Solaris and there your testcase with #define FNM_CASEFOLD
> FNM_IGNORECASE behaves the same as on Linux.
>
>  Jakub

-- 
Markus F.X.J. Oberhumer
oberhumer.com, http://www.oberhumer.com/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fnmatch_test.c
Type: text/x-csrc
Size: 2252 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20040902/49099c66/attachment.bin>