Extras' plague-server and %{ix86}

[Quentin Spencer]

Dan Williams wrote:

>On Tue, 2005-12-27 at 16:36 +0330, Roozbeh Pournader wrote:
>  
>
>>We are having a discussion on 
>>
>>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=176434
>>
>>about whether or not the Extras' plague-server will build i686,
>>i586, ... versions of a package if the ExclusiveArch passed to it is
>>%{ix86}, which according to my /usr/lib/rpm/macros will expand to the
>>following:
>>
>>   i386 i486 i586 i686 pentium3 pentium4 athlon
>>
>>It seems that Thorsten's plague-server actually builds an i686 version
>>of the package, while some packages in extras (like
>>athcool-0.3.11-3.fc5) use the %{ix86} macro and were only built as i386
>>in extras.
>>
>>Would someone with more knowledge of Extras' plague-server enlighten us?
>>    
>>
>
>The buildsystem takes the "base arches" that a particular target can
>build for (usually i386, x86_64, and ppc), grabs the 'buildarchs' RPM
>tag from the SRPM, and filters the buildarchs through the "base arches"
>for the target.  ExcludeArch and ExclusiveArch are taken into account.
>
>It gets more complicated for subarchitectures though.  The buildsystem
>has a feature called "optional arches" that packages can build for if
>the specify it.  That could be set to 'i486 i586 i686', but then you'd
>end up with 4 x86 versions of every package that gets pushed through the
>buildsystem for those people that used %{ix86}.
>
>We clearly don't want 5 different versions of FE, one each for each
>architecture, because Core doesn't do that.  So its unclear to me how to
>proceed with this.  There is another feature called "Additional Package
>Arches" that allows other arches (like i686) on a per-package basis,
>which might be the route to go.  But that feature requires server-side
>configuration updates for each change, meaning that packages can't
>control it.
>  
>

I would like to see something like this. I agree we don't want an 
unnecessary proliferation of packages, but for some packages it is 
justified. I maintain atlas, which is a speed-optimized replacement for 
the blas and lapack math libraries. Currently the package provides 
subarch-optimized libraries using subpackages with names like 
atlas-sse2, but the user has to know about their existence. In the 
future, it would be really nice to have a mechanism to specify 
subarchitectures.

-Quentin