build-from-branch into the primary archive

[Martin Pool]

On 18 February 2011 08:25, Andy Whitcroft <apw at canonical.com> wrote:
> Push to upload implies that it is practicle to move all packages into
> bzr.  For large packages such as the kernel, libreoffice, X etc, where
> those are not in python that would imply we have to extract the intended
> state out of that and then commit it into another VCS just to trigger
> the upload?  I would tend more to think of dput as the 'assembler' way
> of uploading, with pushing to a build branch being the 'best' way to do
> things.
>
> Generally you should not need to force a new way on people.  If the new
> way is better people will switch and use it.  Removing the old way
> should not be needed to achieve your goals should it?

Right, we're not intending to force this on people.  We want to offer
it as an option for people that want to try it out.  If there are
reasons why people find it either doesn't work or isn't a good
tradeoff for some packages, we'd like to work out why and then address
that.

Divergence between branches and uploads is a a problem that crops up
in various places, including the documentation and keeping the package
importer in sync.  Ultimately we hope that developer groups will
choose to say that particular packages must only be uploaded in
branches to avoid people accidentally making things inconsistent, and
we will give them a policy knob to express that.

Things like the kernel where there is substantial upstream-type
development in git are an interesting case.  Possibly we should be
doing the builds there from a git foreign branch.  Making people
manually do a separate commit into bzr might be ok for an early
prototype but I don't think it's a very good solution; fortunately we
already have branch imports etc that we can build on.

-- 
Martin