A need for build triggers & automatic rebuilds

Adam Williamson awilliam at redhat.com
Fri Jun 21 01:11:27 UTC 2013


On Wed, 2013-06-19 at 10:00 -0600, Eric Smith wrote:
> On Wed, Jun 19, 2013 at 4:23 AM, Nicolas Mailhot
> <nicolas.mailhot at laposte.net> wrote:
> > You don't need a special trigger. You need your requirements to match the
> > constraints declared at the rpm level, then incompatibilities will be
> > caught at this level by normal distro tooling.
> 
> How does that work?  A pointer to the docs is fine; My own search
> apparently hasn't turned up the right stuff.

I'm not sure I'd concur with Nicolas that our mechanisms in this area
are entirely sufficient.

If you set up your RPM dependencies to match the actual code
dependencies, then when someone submits an update for one of the
dependencies that breaks the chain, the 'depcheck' AutoQA test ought to
fail, and I think the 'updates-testing report' mail sent to this list
should highlight the issue. But neither of those are 'enforced'
mechanisms: neither would prevent the update being pushed anyway. They
are both advisory. So it would still certainly be the case that this
could wind up out of sync.

Also, putting on my Update Nerd hat for a moment, our current setup not
only *doesn't encourage* coherent Bodhi updates, it more or less
*actively discourages* them. It's very difficult for packagers to edit
each other's updates - even provenpackagers - though pp's can edit
anyone's spec file at the drop of a hat. So if someone else submits an
update for one of Eclipse's dependencies which requires Eclipse to be
updated too, it's very unlikely that the Eclipse maintainers will be
able to do a new Eclipse build and edit it into the same update. Even
though that's how it's supposed to be: updates are supposed to be
internally consistent. If A depends on B, then when B is updated in a
way that breaks A, the Bodhi update should contain builds of both A and
B. There should not be separate Bodhi updates for A and B. But it is
actually very difficult to make this happen, given the way our
permissions system and Bodhi currently work. You have to really
carefully arrange things and have a very high degree of co-operation
between the packagers of A and B: basically, you have to make sure
someone has the necessary rights to be able to submit updates for both A
and B, and no-one else ever submits updates for B, or else even that
person can't fix things. It kind of stinks.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | identi.ca: adamwfedora
http://www.happyassassin.net



More information about the devel mailing list