Triage Working Group Discussion
- Discussed ideas of a triage working group and how that might interact
- Discussed pnkfelix/nikomatsakis taking leave over next few weeks and what to do in the meantime
- Discussed the problems around review queues
pnkfelix will be going on parental leave for 2 months starting July 15. nikomatsakis will be on PTO for a month or so, with limited availability. In the interim, nagisa has volunteered to run meetings.
Another concern was the general review backlog. pnkfelix/nikomatsakis discussed spending a bit of time triaging things that need review and trying to develop a plan.
In general, we noted that it would be useful to keep a calendar on the compiler-team repository to help track this sort of thing.
In general, the compiler triage effort is feeling pretty stretched. It also falls largely on a few people at this point in time. We discussed how we might broaden the effort.
The work of solving a regression
Most errors come in the form of regressions: something that used to compile or run, but no longer does. One observation is that the work of handling a regression is actually something you can break down into many steps:
- Figuring out the impact of the bug and assigns priority
- Coallescing duplicates
- Determining the cause of the regression: we have a tool that bisects to a nightly or even a specific PR, but somebody has to run it
- Minimizing the error case into a standalone example that can be debugged
- Identifying the expected behavior of the test case – sometimes the regression comes from a bug fix!
- Actually fixing the bug
The final two require some amount of expertise – though the precise amount varies. Today, there are a lot of people who help and do the minimization and bisection work, but it would be great if we could help organize those efforts more.
The next thing is one we’ve kicked around for a while – can we create some kind of “ICE-breaker” group that is dedicated to fixing regressions, ICEs, and other “small” bugs in the compiler? What overlap does that group have with the triaging efforts described above?
We discussed this for some time and seemed to come roughly to the following conclusions:
- The #wg-triage effort in the release team obviously overlaps a lot with these efforts.
- Perhaps we should divide responsibility as follows:
- decides what is important to fix or address based on the impact
- helps to bisect, minimize, and identify the source of problems
- brings it to compiler team attention and checks on progress regularly
- tries to find assignees – this will require compiler team participation!
- does the actual fixing
- Note that if wg-triage is going to find out who to assign to a bug, that is going to require compiler team participation. We could in tricky cases have a “liason” who can be assigned that has the job of just reassigning to someone more appropriate.
- On compiler side, the ICE-breakers group might still exist as well, and many of its members would likely participate in the wg-triage effort, but the focus of ICE-breakers would be mostly on the actual fixing of the bug, and less on the bisection and so forth.
We concluded ultimately that we should try to touch base with the wg-triage effort and discuss.
Expert map, reviewing load
We have an expert map on the compiler-team repository, but it is sparsely populated and doesn’t include the new compiler team contributors. That would be in principle be helpful in finding people to assign to fix regressions or to review PRs. We also discused how contributors, even if they don’t know an area well, can still be assigned to do a preliminary review or fix. We need to improve the “expert map” to be more complex and also to include “levels of knowledge”. Perhaps rename it to “knowledge map” or something like that. Also, this can be a great way to figure out things that need to be added to the rustc-dev-guide.
We spent a bit of time discussing how the
I-nominated tags are not
sufficiently precise – at minimum, they should identify the team, but
perhaps also have more information.
Emerging from this meeting:
- Extend expert map with member/contributor levels
- Create a compiler-team PTO calendar where people can note their upcoming absences
- Join discussion with #wg-triage