Re: [RFC PATCH] Optionally use -mlong-double-128 by default on {powerpc{,64},s390{,x},sparc,alpha}-linux

[Jakub Jelinek <jakub at redhat dot com>]

On Fri, Jan 27, 2006 at 12:11:33PM -0500, Andrew Pinski wrote:
> >This is a preliminary patch for -mlong-double-128 as default on the
> >linux targets that are going to switch long double in glibc 2.4+.
> >The gcc side is only done for ppc ATM and I'd say it is not very
> >controversial, just adds a new configure option to select the default
> >and has a reasonable default value (when building against glibc 2.4+
> >uses TFmode long double, otherwise DFmode long double).
> 
> I am still wondering why this was not thought of earlier on in the
> release cycle.  GCC 4.1 is already 9 days overdue and we are talking
> about an ABI change.  Yes this is a good ABI change but we are talking
> about an ABI change which happens a week before release.  In fact a lot
> of -mlong-double-128 support for the targets are new and not as well  
> tested
> as they really could have been if this was changed in say stage 1 of  
> 4.1.

All these changes rely on the C library support, and the support
for backwards compatible way to switch long double has been added to
glibc only a few weeks ago (and on the other side, glibc on several
platforms relies on the GCC side to even build, things have to be
done together).  If the changes aren't done in GCC 4.1 (here I'm talking
mainly about the -mlong-double-128 support, not about the switch to it
by default), then there won't be any released compiler that will be able to
compile glibc 2.4 on ppc32, s390, s390x, alpha nor sparc.
glibc decided to do the long double switchover about a month ago
and glibc 2.4 release is planned within next month or so.

	Jakub