[Discussion] PR reviews

[timokau]

Some incomplete answers:

@tomberek

Could Nixpkgs adopt a policy to ease the update burden? Perhaps a set of tests that would signal to ofborg that the community would accept minor version bumps for specific packages without manual review? This has the cost of increased breakages for people on bleeding edge. But those breakages focus the effort to where it is most critical.

Even if we wanted to do that, how would we automatically differentiate between minor version bumps and potentially malicious PRs?

@tomberek

Increase the number of people who can commit? Assign areas of responsibility?

That has been part of the solution for a while, but its not working very well. I feel like new committers often tend to lose interest / burn out. Its not infinitely scalable either: each committer is a potential security risk for nixpkgs users.

@tomberek

Encourage reviewers/commuters to modify PRs directly instead of relying on the originator review-recommend-edit cycle.

I think that is the exact opposite of what we need for two reasons:

• we have plenty of people who want to get changes in, not many people who review. Shifting burdon from the first group to the second is not going to help
• the people proposing changes are not going to learn that way

@tomberek

Question to reviewers: what slows you down the most? What gives you the most reason the doubt a PR?

Lack of clear commenting of intentions is one thing. Missing upstream reports another. GitHubs UX yet another.

@tomberek

Remove old/stale PRs, or backlog them?

Any other policies that people apply during the review process that could be automated?

Anything the Guix community does that could help?

I still think a process of

• contributor proposes PR, it is tagged with needs:review
• reviewer reviews, tags with needs:changes / merges / tags with needs:merge
• contributor makes necessary changes, re-tags with needs:review.

Then, we can just automatically close all PRs that have been in needs:changes for too long and reviewers have a clear place to look.