On Mon, 30 Jul 2012 18:04:10 -0700, Rob Lanphier robla@wikimedia.org wrote:
Hi everyone,
Y'all know about this, right? :) http://www.mediawiki.org/wiki/Git/Gerrit_evaluation
As I said at the beginning of this process[1], the way this works is we argue for a while, Brion watches, and then he makes the call. You might have also noticed my mention about a meeting today on the subject[2]. That attempted to be a fairly mundane meeting to make sure that Brion had everything he needed to make a decision (and give him the opportunity to say "don't saddle me with this decision") :)
Brion is still on board (yay!) and has everything he needs. Here's the list of alternatives that are under consideration now:
- Staying with Gerrit (with GitHub integration)
- GitHub (full migration, with some mirroring and limited private repos)
- Phabricator
- Gitlab
- Extended evaluation
That last one is one I just added, to make explicit an idea that has been mentioned many times (and one I've articulated myself). Basically, there are many people who are dissatisfied with Gerrit who don't feel they have the time to look into an alternative, but feel strongly that the Wikimedia Foundation should put even more significant resources toward an evaluation.
I'll be blunt, though. Brion made his preference pretty clear that, if he were to make a decision today, it would be to stick with Gerrit, and I've made it pretty clear that's also my bias. The other options (including the "extended evaluation" option) have not been made in a clear and compelling enough way to Brion to sway him. I wanted to make sure that we didn't make the final decision in that room, but it was tough not to call this one early.
However, I was careful to press Brion on what would change his mind in the next two weeks. Basically, it's different for each system:
- For GitHub: we'll need to be convinced that it's ok for our primary
code repository to be hosted elsewhere. That means, in part, a mitigation/backup strategy, better explanation of how account management will work and integrate with our stuff, and we'd need to be convinced we have a strategy for negotiating the license we need for our community. I'd say this one is just Brion's problem, but really, I'm not sure the rest of the org (Erik, our legal dept, etc) would let Brion unilaterally make this call.
- For Phabricator: we've invited the lead developer to the office,
and we'll be hearing what he has to say about the system.
- For Gitlab: Roan would need to make the time to do more of an
evaluation, or someone would have to pick up with what he's started.
- For the "extended evaluation" option, we'd need to hear why it's a
better investment to set up a bunch of alternatives to play around with rather than putting that energy into improving Gerrit, since we likely won't have the time and energy to do both well. We'll also have to understand why we should subject ourselves to even more mailing list debate on this topic :)
We agreed we weren't going to set a hard deadline for how long we're going to stick with whatever decision we make. It of course depends one which alternative we choose. For the "stick with Gerrit" option, a year (more or less) sounds right. Also, despite emphasizing the importance of making a decision and moving forward, we're going to maintain some flexibility still. If we move to Gerrit, we'll be supportive of:
- Pilot projects in other systems
- WMF devs spending 20% time working toward alternatives
- Reevaluating soonish if a really, really compelling alternative
emerges
Rob
[1] July 11 "Gerrit evaluation process" email: http://thread.gmane.org/gmane.science.linguistics.wikipedia.technical/62390 [2] July 26 "Criteria for 'serious alternative'" email: http://thread.gmane.org/gmane.science.linguistics.wikipedia.technical/62648
I didn't do it because I was the one who started the project so it didn't feel right to be the one proposing it myself. But I wonder if I should have put the option of trying to finish Gareth into the mix.