Re: patch: remaining -Wall warnings

[Michael Cook <cook@sightpath.com>]

Ulrich Drepper <drepper@cygnus.com> writes:

> Benjamin Kosnik <bkoz@cygnus.com> writes:
> 
> > I can either break the standard, and have char_traits::eof() not return 
> > int_type but long or unsigned long, 
> > 
> > or
> > 
> > change char_traits::int_type to long or unsigned long
> 
> I don't see the problem.  In the initial mail there was a comparison
> with a negative number.  This never must happen.  So I assume you must
> mean something else.  Defining int_type to wint_t for wide chars
> should be correct.

The char_traits<wchar_t>::eof() value is int_type(-1), which ends up
being the same (in my case, at least) as if `return 0xffffffff' had
been written.

So, I still think the patch I submitted is correct.  But I must
admit I'm not sure I understand what the significance of the -2's in
that code is--in, say, istreambuf_iterator (sbuf_iter.h).  What did
the author really mean when he wrote `_M_c>-2'?  My patch changed
that test to `_M_c!=int_type(-2)' on the assumption that _M_c was
never intended to have a value <-2.

I think -2 means "before the beginning of the file" (i.e., we
haven't yet read anything from this stream).  That would seem to
require that wint_t(-2) is an invalid UCS4 character.  Is that
right?

M.