F12 to require "i686", but which CPUs do not qualify?

[Andrew Haley]

Deji Akingunola wrote:
> On Fri, Aug 14, 2009 at 10:04 AM, Joachim<joachim.frieben at googlemail.com> wrote:
>>> I think there's a valid case for making an exception to this: when a
>>> package is an accelerated version of a particular library.  That is,
>>> when the basic functionality of a library is available in a i686
>>> Fedora package, but a special SSEx version of the library makes use of
>>> faster instructions.

>> Right now, there exist a number of packages which explicitly pull in
>> atlas instead of the also available generic packages blas/lapack which
>> do not exhibit these severe restrictions.
>> Earlier versions of the Fedora atlas package actually supported a
>> wider range of processors including even such offering 3dnow! and also
>> plain x86. The current behaviour (code depending on lapack aborts
>> because of illegal instructions) is a regression which has been
>> introduced by the packager.
>>
> Correction: The current behaviour was not introduced by the packager,
> it is because of changes in the upstream's design of the package;

Yes, we know that it's an upstream change.  I was wondering if there
were some way to configure things so that the library only gets used
when it would work.

> unless of course you mean we should be stuck with the old version.
> The only way to produce atlas binary for architectures not provided
> for in the upstream tarball, is to bootstrap it on that particular
> arch. Unfortunately none of Fedora build infrastructure is based on
> PII or less.

I don't quite understand this.  Why would we need to bootstrap *on* the
old arch to compile it *for* the old arch?  Some configury weirdness,
presumably.  That sounds fixable.  Would it be OK if I did a little
digging to see if I could fix it?

Andrew.