Re: names of distribution-branches in the git repository

[Sylvain Le Gall <gildor@debian.org>]

Hello,

On 22-11-2010, Stéphane Glondu <steph@glondu.net> wrote:
> Le 22/11/2010 09:54, Sylvain Le Gall a écrit :
>> Is experimental/upstream mandatory? I thought it was possible to inject
>> upstream tarball in upstream branch and merge into master (or
>> experimental/master).
>
> Then the upstream of the master branch doesn't correspond to any head?
>
> It is convenient to have a name for the base of the patch series, and
> dom-{apply,save}-patches use that. Committing stuff in the upstream
> branch not meant for master might confuse these tools (and those who are
> used to them :). The upstream branch also plays a role in gbp, so it
> might get confused too. For example, when you don't use pristine-tar,
> which might happen temporarily when you make snapshots, gbp uses that
> branch to generate a tarball. Moreover, branches are cheap in git, and
> cost next to nothing when they are an ancestor of another branch.
>
> In short, experimental/upstream might not be mandatory, but having it is
> convenient and avoids confusion IMHO.

Concerning the role in gbp, it uses branch + tag:

(I disable the pristine-tar)
gbp:info: ocaml-extunix_0.0.1.orig.tar.gz does not exist, creating from 'upstream/0.0.1'

AFAIC, I never touch anything in upstream branch, so I won't be confused
;-)

My point is that upstream branch is upstream and master (or
experimental/master) should refer to upstream + tag. It seems to be the
case and I think it is a consistent choice. 

So even if branches are cheap, I would prefer to have a single upstream
branch. After all we only follow one upstream.

Regards,
Sylvain Le Gall