Copying from #wikimedia-dev IRC (sorry) and cc'ing wikitech-l as requested:
[James_F] cscott: I think you're confused about fab.wmflabs.org. It's not meant for use. It's meant for evaluation.
[cscott] James_F: and i'm saying that unless i can use it for something i can't easily evaluate it
[greg-g] cscott: James_F crazy idea here: can some teams use it for real (I think growth is, kinda?) and export/import to a future real instance? frontend...
[cscott] greg-g: i think that's more or less the plan, but it doesn't really mean that the results will be applicable to the groups who *aren't* using it. that is, i could, for example, use the existing instance for work on the PDF backend, which is a 1-2 person project. but that doesn't mean that the results apply to how the parsoid team works.
[greg-g] cscott: of course, but it can't be on anyone else other than those people to try it and report bugs, no one can impersonate you effectively other than you.
[cscott] greg-g: i totally agree, which is why i'm arguing for a soft transition.
[greg-g] cscott: I think it's mostly semantics at this point, honestly. The difference between a soft transition and the RFC closing with "Yes, pending those blockers are addressed" are the same thing, especially since Platform (the usual suspects saddled with this kind of stuff) won't have time to actually do anything production-like any time soon with Phab.
[James_F] greg-g, cscott: I think that ultimately if we can't work out how we use gerrit and boil it down to 10 bullet points we've failed. It's a simple tool.
[cscott] greg-g: what i'm saying is i don't think it's worth closing the RFC with a "yes, but" since those bugs are all show-stoppers to Real Work right now.
[James_F] greg-g: A "soft transition" means "go make a production Phabricator instance". Which is pretty bad if we never use it properly, and don't shut down other systems.
[gwicke] we could agree to write a new RFC after actually trying it ;)
[YuviPanda] I tried setting our fab.wmflabs.org instance up for doing CR with its own hosted repositories, so we could put a few small projects there spent 3-4 hours, and got close before giving up. I can add other people to the project if they want to give it a shot :) Note that none of this is puppetized so need to be slightly extra careful
[cscott] James_F: if you want to phrase it that way, a "yes but" means, "we'll commit to transitioning to phabricator without ever having seen an instance which will actually work for us"
[greg-g] cscott: how is that different from any development problem/goal?
[cscott] James_F: i think all the extremes are bad. i don't think we should give up on phabricator. and i think it's too early to definitively commit to it. and i think we shouldn't spent 100% of the resources to make a production instance before making a decision, and I don't think we should make the decision without spending *any* of the resources. i don't think we can do a transition without some sort of integration of new and old systems, and i also agree that maintaining the old and new systems together indefinitely defeats the point.
[gwicke] fwiw, I tend to agree with cscott that it's hard to make any informed decision without actually testing it in practice
[James_F] cscott: You're throwing around dramatic terms like "blockers" without explaining what you mean. Again, please take this to wikitech-l and let's have a proper discussion.
[cscott] James_F: I'm arguing for a middle path. devote *some* resources, implement *some* interoperability, decide at *some later* point when we have a more functional instance.
-- [http://cscott.net]