> I think this is wrong.  The type of a single wide character is wint_t
> instead of wchar_t such that it can properly hold WEOF, much in the
> same way as the "normal" character functions use int instead of char.

Thanks for the hint, Mark.  Does the following look correct to you?
I've just tested that it also fixes the issue on AIX, and I got no
regression on amd64-linux.

2009-05-07  Joel Brobecker  <brobecker@adacore.com>

        * c-lang.c (print_wchar): Convert w into a gdb_wchar_t before
        pushing it on the output obstack.

-- 
Joel