Rob Lanphier wrote:
Our Analytics crew have worked out how to generate a graph that gives us a view into our code review backlog: http://gerrit-stats.wmflabs.org/graphs/mediawiki
Thanks for the graph. There are a few typos in the graph's title, by the way. "Mediawiki CodeReview" should be "MediaWiki code review". "MediaWiki" has a capital W and "CodeReview" is a mostly obsolete MediaWiki extension. The graph should also include some explanation text (or at a minimum, a link to your mailing list post). I'm not sure if Limn currently supports this feature, but it should.
As of this writing, there's 207 revisions that have neither positive nor negative reviews associated with them. That's still seems like a pretty big number. 30 of those are more than a month old, and some date back to May.
How is the process working for everyone? Is stuff getting reviewed quickly enough? How has it been for the reviewers out there?
I'm not quite sure why revisions with +1 and +2 are excluded from this graph. This seems to fall into the trap of "code review is the same as code deployment." We've discussed this previously. When people say they want faster code review, they really only mean that as a means to an end (i.e., they really want revisions merged/deployed in a more timely manner). Could a separate graph be created that shows the number of un-merged, un-abandoned revisions with 0, +1, or +2? That seems like it'd be a more accurate picture of the code review backlog.
Generally, I still don't think there's any incentive to clear the queue. If a revision sits for six months, what happens? You wait for someone to "lob a grenade"? :-)
Perhaps revisions that receive no feedback for thirty days should be auto-merged and auto-deployed. The current code review/deployment situation is all based on whim for any code that isn't magically deemed a high priority to the Wikimedia Foundation, though I suppose graphs are as good a place as any to start to resolve that problem.
MZMcBride