Re: fnmatch: rely on more gnulib modules

[Bruno Haible]

Hi Paul,

> Oops, I found a problem when trying to use that patch with coreutils: fnmatch
> now depends on a module wmempcpy that does not exist, causing the bootstrap 
> to fail.

Oops, you can bet that I tested this patch only on glibc systems. :-(
Working on it now...

> Come to think of it, coreutils etc. don't want to worry about thw wmem*
> functions at all, since they don't use any wide-character code. So fnmatch 
> needs
> to be buildable without worrying about providing substitutes for wmempcpy etc.
> And this means the patch needs to be revisited.

But fnmatch needs to work with multibyte characters. When you write "A?Z"
the single character between A and Z may be any multibyte character. Nowadays
UTF-8 locales are very widespread.

fnmatch's implementation is that it converts the pattern and string to wchar_t*
representations (through mbsrtowcs); this is the most reasonable and appropriate
implementation for fnmatch.

You could devise a different implementation that works on the multibyte
representation all the time. It would call mbrtowc many many times and thus
be slower than the current fnmatch implementation.

Bruno