For your consideration: Secondary Architectures in Fedora

[Christopher Aillon]

David Woodhouse wrote:
> On Thu, 2007-05-31 at 19:15 -0400, Christopher Aillon wrote:
>> There is no build for i386 if something fails.  As in any RPMs that may 
>> have been created get deleted, the builds for any unfinished subtasks 
>> are cancelled, and appropriate fail notices sent out to the CLI and via 
>> e-mail.
>> 
>> Doing this any other way will lead to inconsistencies in the primary 
>> repos.  The latest foo would be foo-1.2 on one arch and maybe foo-1.3 on 
>> another.  Having these inconsistencies in the build root will then 
>> become one big mess as nobody would be able to sanely say what version 
>> of foo any given package would be compiled against -- and might cause 
>> more failures if API/ABI changes in the foo package. 
> 
> Nevertheless, this is what's being proposed. Even if a build fails on
> one architecture, it would run to completion on others. 

Not quite.  How do we that right now all of our builds in rawhide don't 
fail on s390 currently?  We don't.  Chances are they work, but maybe 
something in the rawhide compiler broke there.  Assuming the s390 
compiler is broken, would you consider this a partially failed build if 
we haven't started a build on it and therefore can't know the problem 
exists?  It succeeded on i386 and x86-64 and ppc and we push it to the 
repos.

 From a Red Hat perspective, it would be nice if we started doing side 
s390 builds so we can keep on top of any issues as we'll have to care 
about this when we branch for our next Enterprise offering.  If it fails 
in our own side builds, we'd get notified but it wouldn't be a failure 
to the Fedora system because it just didn't start an s390 build at all. 
  But when the Fedora build succeeds, it would be prudent for someone to 
rebuild the s390 package with the same changes.

That is pretty much exactly what is being proposed.  But automated.