Failures of noarch package builds on ARM (Re: ARM Status)

[Daniel P. Berrange]

On Sun, Jul 28, 2013 at 03:05:34PM -0500, Dennis Gilmore wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> hi all,
> 
> I have imported ARM into primary and enabled armv7hl in the arches to
> be built. right now the KDE stack is not entirely built and brought in
> due to https://bugzilla.redhat.com/show_bug.cgi?id=988114 as soon as it
> is resolved we will get everything fixed and brought in. at that point
> arm will be added to the nightly composes. 

I've just attempted to build mingw-libvirt (which is noarch) in rawhide
and was (unlucky?) that it got scheduled on an ARM builder. The build
failed with

  checking build system type... Invalid configuration `armhfp-redhat-linux-gnu': machine `armhfp-redhat' not recognized

  http://koji.fedoraproject.org/koji/taskinfo?taskID=5680000

This is odd because the mingw-libvirt src.rpm has the exact same
tar.gz that we use in the native libvirt src.rpm which accepts
arm as a build system type. So I'd expect that the configure
script is already new enough to work with armv7 hosts at least.

Looking to see if mingw-libvirt was ever tested on ARM secondary
builders, I can't find any build logs. Querying the noarch.rpm
packages hosted on the ARM secondary builder koji host at:

http://arm.koji.fedoraproject.org/packages/mingw-libvirt/1.0.5/1.fc20/noarch/

shows that they were built on the Fedora primary koji instance.

Randomly picking some other mingw packages (mingw-libpng and
mingw-libxml2) shows they were also all build on x86 hosts and
trying a build of them on arm hosts causes the same failure.

So am I right in thinking any noarch RPMs were just copied across
to the ARM koji as-is, without attempting to rebuild them ?

If so, it seems we might have a bunch of noarch packages which
are going to turn out to be broken if unlucky to have their
builds scheduled on ARM.

Side-note, even if the packages worked fine, is it a good idea to
allow noarch builds to be scheduled on ARM, given that ARM is so
much slower ? It seems we're wasting precious ARM builder cycles,
as well as maintainer time, by letting noarch builds go to the
slower ARM builders instead of x86 builder hosts.

/me goes to play russian roulette triggering rebuilds of the
mingw-libvirt package until koji picks an x86 host

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|