Consider PR labels to make reviewing easier

[Jefffrey]

Is your feature request related to a problem or challenge?

Currently we're sitting around ~190 PRs open; would be nice if we can try make reviews easier to try reduce this amount.

Describe the solution you'd like

One thing I can think of is introducing labels such as:

• pending review - Needs someone to take a look at this
• pending author action - Some required changes before we can progress
  • Technically we could use Changes requested review for this but maybe this is easier to see? Or maybe its a softer version, such as asking clarification from author (not necessarily code changes)
• needs discussion - For those PRs which aren't clear cut and need discussion from both sides to progress

This might help reviewers identify which PRs need attention, especially the needs discussion ones as those PRs can last a while and get pushed out in favour of newer PRs (and can lead them to languish); and sometimes they might be marked as draft which can make them harder to see.

A downside is how we apply these labels; more manual action from reviewers probably doesn't help solve the issue at hand 😅

Describe alternatives you've considered

Another thing to consider is making the GitHub action to close stale PRs more aggressive? Though I feel a good number of the open PRs are stale mainly because they haven't been reviewed so might feel unfair to close them aggressively 🤔

Additional context

No response

[Jefffrey]

Would love to hear opinions from others on if there are ways to make identifying & reviewing PRs easier

[2010YOUY01]

I think this is a great idea.

Technically, an open PR should imply pending review and a draft PR should imply pending author action. In practice, though, this is hard to enforce consistently, so explicit labels sound like a good idea.

Proposed PR state labels (from @Jefffrey):

• pending review
• pending author action
• needs discussion

Additional states that could be useful:

• blocked: e.g. waiting for an arrow-rs update before it can proceed
• ready to merge: approved, but according to the policy it's waiting at least 24 hrs before merge
• stalled: pending review for X days and needs attention
• WIP

[alamb]

I agree -- having this labeling would help a lot -- thanks @Jefffrey and @2010YOUY01 for the suggestion

Maybe once we have it working reasonably we could even automate some of the workflows (aka merge after 24 hours, for example)

I personally recommend starting with the smallest number of labels possible and then expand it if we find something that works well. Making a beautiful process that no one actually follows will just make the situation worse I think.

Maybe something like this:

• ready for committer - Needs a maintainer to review
• pending author action - feedback provided, waiting on review
• ready to merge - can be merged after 24 hour window

The "needs discussion" is a tricky one -- is the point of marking a PR like this to try and get it more attention? I sort of worry labeling something "needs discussion" will end up making PRs impossible to merge as it won't be clear when we can restore them.

blocked: e.g. waiting for an arrow-rs update before it can proceed

Note there is already a 'waiting on upstream' label that could be used for this purpose

[Jefffrey]

The "needs discussion" is a tricky one -- is the point of marking a PR like this to try and get it more attention? I sort of worry labeling something "needs discussion" will end up making PRs impossible to merge as it won't be clear when we can restore them.

In my mind its for cases where an author proposes some change which might require larger changes, or just wants to see if they're going in the right direction; such PRs would usually be marked draft because they aren't ready for merge but might be overlooked as being interpreted as the author themselves still refining the PR.

[alamb]

In my mind its for cases where an author proposes some change which might require larger changes, or just wants to see if they're going in the right direction; such PRs would usually be marked draft because they aren't ready for merge but might be overlooked as being interpreted as the author themselves still refining the PR.

I do agree it would be amazing to find some way to filter for "needs attention" but doesn't also become such a wide net that everything gets marked that way 🤔

[alamb]

Maybe this is related to the discussion about "improvement proposals" (which was also about helping gather feedback for larger changes) 🤔

• Discussion: DataFusion Improvement Proposal (DIPs) Process? #16886