Re: we now have "lilypond" organization on GitHub

[David Kastrup]

Urs Liska <address@hidden> writes:

> Am 23.09.2013 02:55, schrieb Graham Percival:
>> On Sun, Sep 22, 2013 at 02:57:42PM +0200, Urs Liska wrote:
>>> Am 16.09.2013 12:50, schrieb David Kastrup:
>>>> So the question is what we should be telling the Savannah operators to
>>>> make working on GNU projects using Git more feasible.
>>>>
>>> What about asking them to provide Gerrit as  a service?
>> That's a possibility worth investigating.
>>
>>> - The core developers have the right to approve/reject proposals
>>>    as well as pushing directly to the main repo
>>> - Approval of a patch immediately merges it into the main code base.
>> I think you mean "push directly to staging".  We don't want *any*
>> patches going directly to master without checking that they still
>> compile.
>
> I didn't elaborate on this because staging is by default integrated in
> the Gerrit workflow.
> Its standard setup
> http://gerrit-documentation.googlecode.com/svn/Documentation/2.7/intro-quick.html
> is having two stages:
>
> a) Code review
> Reviewers rate a patch with -2 up to +2 while the +/-1 are opinions.
> A suggested patch passes that stage with at least on +2 and _no_ -2
>
> b) Verification
> This is suggested to be done automatically through build-tests or
> whatever is appropriate for the project, or semiautomatically.
> A patch that is 'verfied' will then be merged automatically from
> Gerrit to the main repository.
>
> The process can be configured to require less or more
> tests. E.g. removing the verification from Gerrit because it is done
> somewhere else.

We'll definitely need the verification be done externally: even if
Savannah were up to carrying the load of our pre-commit testing, we want
to have it under our own supervision.

> PS (regarding Graham's other mail): The intention of suggesting Gerrit
> is to keep development where it currently is but streamlining
> contribution, especially for non-members.

It would be actually a rather small step: replacing Rietveld with
something allegedly better catering to Git, hopefully facilitating the
review of commit series rather than single commits.

Quite a few of our most troublesome commits are of the "dozen unrelated
things squashed into a single commit" kind encouraged by our use of
Rietveld, and that makes pinpointing and fixing problems selectively
very hard.

It may be that something like Gitorious would obsolete Gerrit (as well
as the Google issue tracker), but then we need to start somewhere.

If there are no objections in the next few days, I'll try bringing the
topic up on the Savannah hackers list.

-- 
David Kastrup