voting in gerrit: merge at +3? (+2+1 / +1+1+1)

Stefan Sperling stsp at stsp.name
Tue Nov 13 15:43:50 UTC 2018


There has been some face-to-face discussion about Osmocom's code review
process within sysmocom recently. I am posting to this list now with the
consent of everyone involved so far, in order to involve the Osmocom
community at large in this discussion as well.

There has been a lack of code review from people who don't have "+2"
super powers in Gerrit. This applies to anyone among us, independently
of any individual's relationship with sysmocom. The bulk of the work
involved in reviewing code falls on Harald's shoulders, with Pau and
Neels sharing most of the rest between themselves.

At present, while people who add a +1 have their voices heard, their
input does not formally affect the decision to merge a change.
A change still has to pass by the pre-selected +2 gatekeepers in order
to be merged, no matter how many other people have provided input.

The implication for developers without +2 powers is that their time
is more effectively spent on advancing their own changes towards
a +2 vote, rather than spending time on whatever else is waiting
in Gerrit. This may not apply to everyone, of course. But at least
for me, this is certainly the case; I have only been reviewing other
people's patches when I was explicitly asked to do so.

Myself and several other developers hope that with a change to our review
process we can fix this inertia, spread code review across more shoulders
and encourage more collaborative code review in Osmocom.

The basic idea is that everyone's input should count for something.
If those among us with +1 powers were given a partial say on the fate
of every change, our decisions will carry more weight and our influence
within the project will slightly increase. We'd also be encouraged to step
out of our own corners of expertise every now and then, and look at what
other developers are working on. On the flip side, this means we'd carry
more responsibility than we do now. We wouldn't always be relying on our
+2 gate keepers and would have to apply our own judgement more carefully.

The concrete proposal is to make votes in Gerrit accumulative.
Each change would require a total score of +3 to be merged. This score can
consist of either a +2 and a +1 vote, or three +1 votes; and no -1 votes.
Also, +2 developers would keep their ability to unilaterally block or revert
changes under this new model. They'd keep their existing role as arbiters
in case of disputes.

Max figured out how Gerrit could be configured for this behaviour.
It involves Prolog code, but since we're all quite smart we should be able
to figure that out, right?
https://gerrit-review.googlesource.com/Documentation/prolog-cookbook.html#_example_13_1_1_2_code_review

We'd be interested to hear what the community thinks about this proposal.

Thanks,
Stefan


More information about the OpenBSC mailing list