build-from-branch into the primary archive

[Scott Kitterman]

On Tuesday, February 22, 2011 01:18:26 am Steve Langasek wrote:
> On Tue, Feb 22, 2011 at 01:01:45AM -0500, Scott Kitterman wrote:
> > On Tuesday, February 22, 2011 12:47:15 am Steve Langasek wrote:
> > > On Tue, Feb 22, 2011 at 03:57:16PM +1100, Martin Pool wrote:
> > > > Regarding where it is done, I see no problem with doing it in
> > > > debian/control.  If it's configured in the package itself we would
> > > > have the option to give a warning at the time people run dput rather
> > > > than later sending mail back from Soyuz complaining about it.
> > > 
> > > If you're going to put it in debian/control, then I think a Vcs-Bzr:
> > > field pointing at a UDD branch already encodes this information -
> > > 'apt-get source' already warns about it, we might as well have dput
> > > warn on the same thing.
> > 
> > Apt-get source already warns on any vcs fields (including Debian ones),
> > so I think if you're going to depend on those, then it won't be very
> > effective.  I think the idea of some kind of UDD specific 'are you sure'
> > sort of option is reasonable.  Failing to consult non-UDD branches
> > doesn't have the same effect.  It may result in things getting out of
> > sync, but it doesn't result in history in the VCS getting over-written.
> 
> I think if a Vcs-Bzr field points at a UDD branch, that *is* UDD specific.
> The information is already there; there's no reason to encode the same
> information in a different field, and unlike for Vcs* fields generally, if
> it points at a UDD branch in Launchpad, we have a high degree of certainty
> that we're not looking at stale metadata from Debian.
> 
> It's straightforward to have the 'are you sure' option check specifically
> for UDD branches in this field.
> 
> > P.S. I see those warning all the time and I confess I almost never look
> > at the VCS.  I don't recall it having been a poblem and I don't expect
> > I'll change.  I suspect I'm not alone in this.
> 
> I often leave staged changes in UDD branches of packages that I'm working
> on (not for very long, unless the Vcs-Bzr field *is* marked).  It
> invariably makes more work for me to reconcile the branch with the archive
> after the fact than it would've taken someone else to commit to the branch
> in the first place.  I call that a problem.

Right.  I meant it hadn't been a problem for me.  I recognize that experience 
is not universal.

One point I don't understand is why people insist they need to leave work in 
progress on the official branch?  bzr is a DVCS, so why don't people make 
their own branch and then only push to the official branch when it is, in 
fact, ready for upload.  If we did it that way, then pushing to the official 
branch could be limited to people who could upload the package, build from 
branch could be triggered by the push, and there'd be no problem with dput 
uploads overwriting work in progress.

Scott K