On Sat, Apr 17, 2010 at 04:29:06PM +0100, Nicholas Clark wrote: > > > On Wed, 14 Apr 2010 21:07:38 +0300, Niko Tyni <ntyni@debian.org> wrote: > > > > we have a long-standing request [1] to set -Duse64bitint on the 32 bit > > > > Debian architectures. Now that there's going to be an ABI change with > > > > 5.12.0 this would be possible, but I'm wondering if it's the right thing > > > > to do. > You won't please everyone, whatever you do. Thanks for the replies and particularly for the concise summary :) To conclude, I intend to set -Duse64bitint but not -Duselongdouble on all Debian architectures to aim for consistency between them as far as reasonably possible. We've got such a wide range of long double implementations (64, 96, and 128 bits and one that's not even IEEE 754 compliant) that using them for NVs would surely lead to unnoticed accuracy problems on the less used architectures. We've had NVs with less precision than IVs on the 64-bit architectures for ages, so I don't think it's going to be a big issue for the 32-bit architectures either. See also the brief discussion at http://lists.debian.org/debian-devel/2010/05/msg00078.html (but please note that the subthread about "pure" 32-bit architectures is mostly a misunderstanding.) BTW, this has unearthed two apparent gcc 4.4 bugs, one on arm and the other on sparc. For details, see http://bugs.debian.org/580334 http://bugs.debian.org/577016 -- Niko Tyni ntyni@debian.orgThread Previous