Re: Eclipse EMF build failures on ppc64 hosts

[Adam Williamson <awilliam redhat com>]

On Fri, 2009-03-06 at 08:14 +0000, Mat Booth wrote:

> > We looked into this a bit today, and it appears to be a shortfall in
> > mock, or rpm(build) that is triggered by koji.  When koji builds
> > something for noarch, it explicitly passes --target noarch.  This
> > overwrites some settings in mock for the rpmbuild arch and in the ppc64
> > case, rpm defaults to ppc, and libdir is wrong.  If koji didn't pass
> > --target, then the default arch for the config would be used and all is
> > happy.  We're not entirely certain where the real bug lies, but we have
> > a couple possible work arounds, and we'll be discussing with rpm folks
> > tomorrow.
> >
> 
> Thanks for the feedback, Jesse. Much appreciated.
> 
> You've probably seen me raise a rel-eng ticket about this already so
> the problem doesn't get forgotten.
> 
> In the meantime, I've hammered away of the resubmit button in Koji and
> my package has now built successfully.

Um. Is there really a bug here?

I thought Mamoru had it correct, and the bug was in the spec: it
declared noarch but used %_libdir, which is not valid and makes no
sense, %_libdir being inherently arch-dependent. It shouldn't be
expected for %_libdir to 'work' in a noarch build, AFAICT, and if
anything, the behaviour of always expanding to /usr/lib for a noarch
build is probably the most sensible thing to do. 
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Fedora Talk: adamwill AT fedoraproject DOT org
http://www.happyassassin.net