Debian Sync - Re: Let's Discuss Interim Releases (and a Rolling Release)

[Colin Watson]

On Thu, Feb 28, 2013 at 12:59:19PM -0800, Steve Langasek wrote:
> On Thu, Feb 28, 2013 at 03:11:27PM -0500, Jeremy Bicha wrote:
> > I think we need to train our britney to block on Debian or Ubuntu RC
> > bugs. Maybe this will also allow the Kubuntu developers to package the
> > KDE beta updates without needing to worry about those getting picked
> > up in the next (monthly?) update cycle.
> 
> It is fundamental to the model that has been implemented for $devel-proposed
> in Ubuntu that we *don't* block packages in -proposed for anything other
> than consistency and installability, because to do otherwise would
> dramatically increase the on-hands management required to keep -proposed
> from becoming a tangled logjam.  We don't want to reproduce that part of the
> Debian testing experience.

I agree with most of this (and perhaps this is something that only has a
visceral appeal if you've been a Debian release manager ...).

That said, the worst problem is the time delay.  I didn't make our
proposed-migration instance check RC bugs because it didn't seem
desperately useful since users wouldn't have a chance to file them.  But
developers often have some idea that something might be risky when they
upload it too, or perhaps realise shortly afterwards, and even without
introducing an artificial delay there's a window while everything
builds.

Perhaps it would be useful to check Ubuntu Critical-priority bugs
anyway, without a delay?  This would have two advantages:

 1) The process for handling "argh, my upload was broken" goes from
    "find a member of ubuntu-release and get them to commit a block
    hint" to "file a critical bug".

 2) There would be an incentive for the Critical priority to actually
    mean something for Ubuntu.  We aren't a million miles away from it
    being useful, but there's never been a real incentive to clean
    things up all the way.

Debian RC bugs are trickier.  Firstly, you'd have to check versions too.
Secondly, RC bugs are often due to more than just the version of the
specific package they're filed against, so we'd end up with a lot of
exceptions.  I can see why it's tempting but I think there may be too
many problems on the flip side for it to be worth it.

-- 
Colin Watson                                       [cjwatson at ubuntu.com]