Re: mon_grouping bug again

[Geoff Keating <geoffk at geoffk dot org>]

Roland McGrath <roland at redhat dot com> writes:

> > http://sources.redhat.com/ml/libc-alpha/2003-03/msg00451.html
> > 
> > Could someone comment on how to proceed with this issue.
> 
> This patch is about right.  It is necessary for correct behavior according
> to the spec that the values from localeconv use the CHAR_MAX value seen by
> programs using <limits.h>.  The only reason I have not put it in yet is
> concern about the possible effect of the change on existing binaries on
> unsigned char platforms that might have been built to work around the
> problem.  Since noone representing the affected platforms (arm, ppc, s390,
> s390x) has spoken up, I assume it as at least not known to be an issue.
> It's a pretty minor breakage if it happens.  
> 
> I am also vaguely concerned about people using -fsigned-char, but I don't
> see a reasonable way around that.  (It's technically possible, using
> separate ABI entry points chosen at compile-time depending on
> __CHAR_UNSIGNED__; but that is too ugly to contemplate doing.)  
> On platforms where char defaults to unsigned char, using -fsigned-char can
> break the standard C environment.

On powerpc, similar issues are known (there are a few other places in
the locale specification where CHAR_MAX means something), and the
advice I give people is "if this bites you, it's time to stop using
-fsigned-char".

-- 
- Geoffrey Keating <geoffk at geoffk dot org>