add bors to CI

[davidhewitt]

Sounds like we are generally in favour of doing this.

@messense - are you planning to do this? Otherwise I might try to have a play next in the next week or so.

[adamreichold]

So which jobs are we running on each PR before running all of them on the main, staging and trying branch.

I'd suggest fmt, clippy, build (stable Rust on all OS, MSRV and nightly on Linux only, latest stable Python only or at most all Python versions on stable Rust and Linux only), and coverage (one OS).

@davidhewitt I just noticed that our 32-bit Windows build still targets CPython 3.10 instead of .11. Is that intentional?

EDIT: Actually, all include: matrix jobs are still on CPython 3.10.

[adamreichold]

build (stable Rust on all OS, MSRV and nightly on Linux only, latest stable Python only or at most all Python versions on stable Rust and Linux only), and coverage (one OS).

Theoretically going even more minimal, e.g. running the tests for coverage once might also be useful since we will still test the full matrix before merging into main.

[messense]

are you planning to do this? Otherwise I might try to have a play next in the next week or so.

I'm currently working on PyO3/maturin#1221 and PyO3/maturin#1214, please go ahead if you have time.

[davidhewitt]

Theoretically going even more minimal, e.g. running the tests for coverage once might also be useful since we will still test the full matrix before merging into main.

Yes, I'd be tempted to keep the on-push checks as light as possible, something like run check (not test) for each OS / Python version, and then one run of the test suite on each OS w. latest Python? Should catch most issues and then full regression test run can make sure everything is tip-top shape.

[adamreichold]

one run of the test suite on each OS w. latest Python?

Sounds reasonable. I would argue to do this with coverage though as this is information we need before approving.

[davidhewitt]

Agreed

[messense]

Build cache will be a problem for pull requests, GitHub Actions cache has restrictions for accessing a cache, PR usually can only access caches from default main branch, but if we don't run workflows on main there will be no cache available.