Update pushing and bugzilla workflow

[Kevin Kofler]

Andrew Lutomirski wrote:
> But now I want to submit fc22 and fc23 updates.  The only option I'm
> aware of is to submit two separate updates in bodhi.  In the process,
> I have to choose whether to ask bodhi to auto-close the bugzilla bug.
> Both options are wrong, unfortunately.

I pick the "most appropriate" release (e.g., the one against which the 
update was filed), mark the bug as auto-close there, and keep the auto-close 
checkbox unchecked for the other releases.

> Furthermore, now that I have two updates pending, I have a dependency.
> If I want to be a good packaging citizen, I shouldn't mark the fc22
> update stable before marking the fc23 update stable, because doing
> that breaks the update path.  I also don't want to close the bugzilla
> bug until *both* updates are stable.

I always disable autokarma, and I'll queue the updates to stable at the same 
time. In freeze periods, I will in most cases NOT push the updates to the 
stable releases to stable before the branched frozen release is unfrozen, 
because doing so would necessarily break the upgrade path.

If we are in early branched phase where the branched release has a shorter 
timeout (3 days vs. 7), what I typically do is that I don't pick Branched as 
the release that will auto-close the bug, but one of the stable releases. 
That way I can push Branched after only 3 days and the bug will stay open 
until the stable releases get the fix. It also does not break upgrade paths 
to push Branched earlier. (It's doing the opposite that would be broken.)

> AutoQA and taskotron are basically useless, because the breakage only
> occurs if the fc22 update is marked stable before the fc23 update, and
> taskotron is done before that's even possible.

I do not trust any of those tools, I just take care of the upgrade path by 
hand.

> Would it make sense to have a composite update type that targets
> multiple releases, where (a) newer targeted releases must be marked
> stable before older targeted releases and (b) the bugzilla bug is
> closed once all of the targetted releases are pushed?

Yes, but most importantly the karma should be shared across all releases so 
that they can be queued all at once even if (as usual) nobody bothered 
testing Fedora n-1.

        Kevin Kofler