[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [tycho-dev] gerrit

See inline.

> -----Original Message-----
> From: tycho-dev-bounces@xxxxxxxxxxx [mailto:tycho-dev-bounces@xxxxxxxxxxx]
> On Behalf Of Igor Fedorenko
> Sent: Montag, 7. Mai 2012 16:42
> To: Tycho developers list
> Subject: [tycho-dev] gerrit
> 
> I have a question/problem with our current Gerrit workflow I'd like to
> discuss...
> 
> First, I find overhead of using Gerrit very high. We have to stage the
> change, give other developers time to review, rebase, stage rebased
> change and finally merge. Having to rebase after wait pretty much forces
> development of each change on a feature branch, which adds to the
> overall overhead.

I find the "rebase, propose rebased & submit" steps at the end acceptable when I do them in one shot, in particular without waiting for the voter job. Thinking about it, these steps can be further sped up by pushing the rebased change directly to master. Gerrit will notice this and auto-close the review for the change. So the overall process would be:

1. Propose a change by pushing it to refs/for/master
2. Give others the chance to review (for about 2-3 days from the first proposal)
3. Rebase the change and push it directly to refs/heads/master. This will auto-close the change created in step 1

About the feature branches: I develop all non-experimental changes in master and propose them one by one (so that they depend on each other in Gerrit). Since they are in most cases also submitted in the same order, I don't need any feature branches locally...

> Second, I actually don't see much value in doing code reviews for
> majority of changes given Tycho dev team size and composition, i.e. very
> small number of developers with about the same level of experience with
> the code. All review comments I made or received, at least so far, were
> minor and I believe the same discussion can happen without Gerrit.

I actually feel that Gerrit is a huge step forward when it comes to talking about code. I don't see that this is really possible through any other means. And hence by not using Gerrit, I think that we would deprive us of the possibility to talk about the code.

> So my proposal is to push directly to master as the normal course of
> action and use Gerrit reviews only when we want to ask other developers
> opinion/comments. In the worst case we'd need to go back and
> revert/rework some of the commits that went to master directly, but I
> expect this to happen very rarely.
> 
> What do you think?

I don't think that we should give up on Gerrit already, but rather try to make it work for us. IMHO the benefits outweigh the costs, and there is room for improvement (see above).

Best regards
Tobias


> --
> Regards,
> Igor
> _______________________________________________
> tycho-dev mailing list
> tycho-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/tycho-dev