Public Meeting, Tuesday March 2 2021

The next Numba public meeting is Tuesday 2 March 2021, same details as usual.

Items for this meeting include:

  • Numba 0.53 RC2 and Final release plan
  • Numba governance & community: redefining roles

Suggestions for discussion topics are welcomed, please do comment below.

If you are interested, please do join and be part of the discussion. Everyone is welcome!

Joining details at: Weekly Open Dev Meeting 2021 .

There are two follow-ons from things said in the meeting that I’d like to add:

A note of thanks

There was some discussion of why it’s difficult to get into working on Numba related to the complexity of its internals, and various suggestions as to what specifically makes it difficult (including by myself, around the complexity of the structure of the code when stepping through working out what’s going on). Stuart referred to the amount of effort required to refactor the code, and reminded us that there’s already been a lot of exhausting effort gone in to improve the core of Numba to make it more logical, coherent, and extensible.

I’d like to express my appreciation for this work - it has made real and tangible benefits to the core of Numba for extending it and for its comprehensibility - the ease with which I could recently add a CUDA-specific pipeline (Add CUDA-specific pipeline by gmarkall · Pull Request #6731 · numba/numba · GitHub) and re-use more of the Codegen and Codelibrary abstractions (CUDA: Replace `CachedPTX` and `CachedCUFunction` with `CUDACodeLibrary` functionality by gmarkall · Pull Request #6769 · numba/numba · GitHub) in addition to reusing the Dispatcher mechanism for CUDA (https://github.com/numba/numba/pull/6432) is a testament to this.

So: not only do I appreciate the improvements that the refactoring brought, I’d also like to publicly and personally thank Stuart for the large, sustained effort that he put into reorganising and tidying up all the abstractions in Numba over the past 18 months. Thank you, @stuartarchibald!

An offer to mentor triagers

I’ve been in the triager role for almost a year. As I mentioned in the meeting, fulfilling this role is a great way to explore many parts of Numba that I wouldn’t otherwise reached into solely by picking bugs and issues and working through solving them. The process of checking reproducers and looking into the difficulty of the fix (particularly spending a few minutes to see if a quick fix is possible) brings familiarity with many little pieces of Numba - in isolation these don’t mean much at first, but with enough issue investigations the dots begin to connect and give an overall picture of how things work.

It sounded like there are people who are keen to get into contributing to Numba (or to encourage and support others to do so), who might be interested in doing triaging, but would prefer to begin with some support. To this end, I’m offering to mentor new triagers. Concretely, this offer would mean I’d:

  • Before a new triager’s first week on duty, to walk them through the process, share my experience, and answer questions.
  • During their first few sessions on duty (I would imagine their 1-week sessions would be 4-5 weeks apart, inferred from the current cadence of on/off duty that we have for triaging), I’d provide backup to the new triager - they would go about the usual triage duties, and any issues / PRs / gitter threads / Discourse posts where they feel they need some advice and guidance, I’d provide advice on.

Next steps

I understand the discussion about governance will continue on to the next public meeting on March 9th - I’m happy to discuss the above more with anyone interested there or through public channels prior to the meeting. If anyone would like to ask questions or discuss triaging with me privately, I can be contacted through a private chat on Gitter (I’m @gmarkall there) or via email at gmarkall@nvidia.com.

2 Likes

Hi @gmarkall,

Re: note of thanks

Just wanted to say many thanks for your kind words :slight_smile: I’m pleased to hear that all the effort has made a difference!

Whilst I’ve undertaken quite a lot of the larger scale refactoring, most of the Numba core developers have also taken part, and it would also not have been possible without the community testers, so my thanks go to all those involved!

I’m pleased to hear that the pipeline work went well for CUDA, the design is mostly worked out now but there are still some rough edges that could do with improvement. I’m also hoping that the current refactoring efforts around @overload will massively help with Numba’s extensibility… watch this space :slight_smile:

Re: triage mentoring

Thanks for offering to do this, the approach you suggest seems good and perhaps something that can be added to the wiki, maybe in/next to the czar cheatsheet? Am looking forward to continuing this discussion in the public meeting.