http://fedoraproject.org/wiki/TomCallaway/SecondaryArchitectures

[David Woodhouse]

On Tue, 2007-07-10 at 14:42 -0500, Tom "spot" Callaway wrote:
> On Tue, 2007-07-10 at 12:05 -0500, Josh Boyer wrote:
> 
> > > I'm only suggesting that the packager should need to _look_ at the
> > > failure before filing the required ExcludeArch bug and pushing a 'ship
> > > it anyway' button. I'm not saying that we should force them to start the
> > > build from scratch again like we do at the moment.
> > 
> > I agree.  This is the largest objection I have to the current proposal.
> 
> This is a significant workflow change.

It's a change from our current 'failure causes the whole build to abort'
situation, yes. That's probably a good thing though.

> How should we force the packager
> to look at the failure before letting the packages push? 

I assume it'd just be another command to koji; from the command line or
through the web interface. To push a partially-failed build, you just
give it the bug number as required for the ExcludeArch, it'll (probably)
rebuild the SRPM with the ExcludeArch added and then release the binary
versions which _did_ build into their respective repositories.

That doesn't seem like such a divergence from what we have a the moment,
which waits for a final 'OK' that the package has built OK everywhere,
before releasing the build to rawhide. It's just that it would now be
"built OK _or_ approved anyway" instead of just "built OK".

If a partially-failed build gets pushed, we may have to make the
architectures on which it failed use the _older_ version of the package,
rather than losing the package from the repository entirely. Otherwise,
you end up with thinks like kernel-headers going missing and screwing
the whole system whenever a kernel fails to build. Things _used_ to work
like this, until relatively recently.

-- 
dwmw2