Guidance on triaging open compiler issues

Development Compiler
1

I'm a relatively new member of the swiftlang GitHub organization and the swift-committers team. I'd love to start helping triage issues on swiftlang/swift, but wanted to check in first.

For context, most of my contributions so far have been to swift-subprocess, where I have about 40 merged PRs. I've also made a smaller number of contributions to the compiler and swift-syntax, in the form of issues and PRs:

There are currently about 1,200 open compiler issues needing triage. Would it be appropriate for me to start triaging some of these by adding labels and, where I'm confident, suggesting possible workarounds?

If so, is there any guidance existing triagers, maintainers, and other contributors would give? For example, I have a few general questions:

  • Should I prioritize older issues, newer ones, or just whichever issues I feel most confident about?
  • Would it be better to add the labels I think apply, but also leave "triage needed" in place until someone more senior reviews it?
  • What's the preferred approach to closing issues? For example, if I suspect an issue is a duplicate, my initial approach would be to add the necessary labels and comment with the potential duplicate issue to link them, and then wait to see if the issue author closes it themselves or provides some additional context. If they don't, is there a preferred time period to wait before closing an issue?

Please feel free to share any other information that may be helpful. Thanks!

3 Likes
2

@anthonylatsis seems the most active on these matters for the compiler issues, he probably has some answers.

1 Like
3

Aren't these mutually exclusive?

1 Like
4

I don't think so. swiftlang/swift is a major repo with a ton of activity, and the number of open issues/PRs is comparable to other major repos.

Like @Finagolfin mentioned, there are existing triagers working through the backlog. Before gaining write access, I "unofficially" triaged a handful of issues by leaving duplicate comments and suggested workarounds, and on some of these issues @anthonylatsis later came by to add the necessary labels.

There's so much to do and so little time, which is why I'd love to get started helping out!

2 Likes
5

Has anyone experimented with LLMs or ML to help automate the triage bottlenecks?

6

I think @jamieQ did at some point with decent success.

Edit: Some questions on repo contributor culture & etiquette

1 Like
7

Thanks for sharing this post! I actually read it when it was posted, but forgot about it. Based on that thread and this one, I think I'll take a generally conservative approach for now. I'll continue monitoring these threads and any similar ones that come up, and also continue to learn from how I see others triage issues.