This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] change GLIBC PPC64/ELF2 ABI default to 2.17
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Adam Conrad <adconrad at 0c3 dot net>
- Cc: Carlos O'Donell <carlos at redhat dot com>, <munroesj at us dot ibm dot com>, <libc-alpha at sourceware dot org>, Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>, Jeff Law <law at redhat dot com>
- Date: Wed, 29 Jan 2014 18:57:31 +0000
- Subject: Re: [PATCH] change GLIBC PPC64/ELF2 ABI default to 2.17
- Authentication-results: sourceware.org; auth=none
- References: <1391008726 dot 16702 dot 105 dot camel at spokane1 dot rchland dot ibm dot com> <52E92E7C dot 1040707 at redhat dot com> <20140129172158 dot GT15976 at 0c3 dot net>
On Wed, 29 Jan 2014, Adam Conrad wrote:
> > Not withstanding a double-double to IEEE binary128 transition which
> > the entire community would like to see happen as double-double is
> > very difficult to support and maintain.
>
> This double-double transition rumour is something it would be nice to
> nail down as well, especially if it's something we can sort earlier
> than later, but it's also not as world-ending an ABI break as changing
> the baseline symbol version.
My position on that is that it should be considered a new ABI, with symbol
versions set accordingly to the next glibc release after support is added;
that supporting multiple long double variants, neither equivalent to
double, in a single glibc build would be significantly worse than the
existing compatibility support for configurations where long double
changed from being equivalent to double. (I don't rule out a merge of
libquadmath support into glibc - which would mean two different
floating-point formats with 16-byte size on x86_64 - given sufficient
interest in TS 18661-3 support. But that would involve different function
names, e.g. cosl and cosf128, which is much less problematic than allowing
for the same name possibly meaning two different functions.)
(A patch could of course demonstrate that support for two long double
variants isn't as bad as I expect, but my starting assumption is as
above.)
The existing work on binary128 on ppc64 is on
svn://gcc.gnu.org/svn/gcc/branches/ibm/gcc-4_9-ieee/.
--
Joseph S. Myers
joseph@codesourcery.com