Contributing to jax

[JanLuca]

For me as a unregular contributor to jax, I find the review process a little bit unclear. I know that reviewing pull requests takes some time and that a response will not come right back. But viewing on the project from outside, I got the impression that pull requests by non-core devs are sometimes lost in the flood of automatic PRs created for example by tools as copybara-service.

My problem is that there are no guidelines on what to expect for the review process. Since there is also no overview of who might be the right person responsible for certain code parts, you cannot just ping people politely to ask if they can take a look at it.

I don't want this to be a criticism of anyone (I really like the work of the core devs and all the features of jax!), but simply as an invitation for feedback on how the contribution is perceived from the outside.

[jakevdp]

Hi - sorry about the lack of response for your PR (#32880)! In this case it's likely due to the fact that your PR is very hard to review, and nobody on the team really has the expertise to evaluate it (it's 1400+ lines of very domain-specific logic). Still, we should have responded sooner, and I'm sorry for that.

To be frank, we've started to receive more and more AI contributions via pull requests, and these are typically large changes that come out of the blue without any context (much like your PR) and so it's hard to give the attention to these changes that we would have a few years ago.

In the future, I'd suggest floating ideas via the issue or discussion tracker before spending significant time on an implementation. Code contributions like this are not just a request to merge the PR, but a request for our team to support and maintain that code indefinitely: taking on that support burden for complicated code is not a small request.

If you had asked before writing the code, we would have pointed you to https://docs.jax.dev/en/latest/jep/18137-numpy-scipy-scope.html#scipy-sparse, where we decided previously to not prioritize new work on jax.scipy.sparse.

[JanLuca]

Thanks for the long response. As said, I do not want to blame anyone. I know this is a big project and as said, I like to contribute occasionally since I like the work :) And I see that the review burden on the core team side increased due to AI contributions.

For context of my contributions: I am the main developer of a Physics' library called "variPEPS" which is built around jax.

I just checked the contribution guideline and I really overlooked the part

For other proposals, we ask that you first open a GitHub Issue or Discussion to seek feedback on your planned contribution.

Sorry for not living up to that.