Set up CI for wasm32-emscripten target

[hoodmane]

This adds CI tests for wasm32-emscripten target. CI run builds libpython3.11.a for the wasm32-emscripten target and then runs cargo test against it.
Resolves #2412.

• fix or xfail broken tests

[hoodmane]

So the segfault seems to be because PyCapsule_Import("datetime.datetime_CAPI") returns NULL.

[hoodmane]

It would be good to return an error from PyDateTime_IMPORT if the import fails.

[davidhewitt]

Thanks, will try to give full review later!

It would be good to return an error from PyDateTime_IMPORT if the import fails.

We shouldn't change the ffi symbol, however in ensure_datetime_api in types/datetime.rs we should definitely assert the API is non-null after import.

Maybe worth correcting as a separate PR and documenting as a fix in the CHANGELOG.

[hoodmane]

Maybe worth correcting as a separate PR and documenting as a fix in the CHANGELOG.

Okay I will open a pr. I guess I should add a changelog entry here too.

[davidhewitt]

This is really neat; I'm very excited to see this added to our CI. Thank you so much for putting the time in for this! (I really need to get my head around how shared libraries work in wasm sometime.)

I guess I should add a changelog entry here too.

As strictly speaking there's not anything user-facing here I don't think a CHANGELOG is necessary for this one. Any subsequent MRs where we fix issues with wasm which users will encounter would be worth documenting. (e.g. the datetime fix.)

I've added a few comments - I think we can drop the instant patch, and also I'd like to have ignore = "..." notes on tests which we know will never work.

Let's keep #2412 open after this merges, and then the remaining tests which we believe should work but currently don't can be tracked there. Hopefully we can close everything off before the 0.17 release. (I've still got a couple of axes of feature work I'd like to finish first, so there's some time!)

[mejrs]

I'd like to have ignore = "..." notes on tests which we know will never work.

I've always used #[ignore] to disable tests that are very expensive to run, rather than to not run tests that don't work at all. After all, what is the point of cargo test -- --ignored then? So I would personally prefer:

#[cfg(not(target_arch = "wasm32"))]
#[test]
fn test() {
    // ...
}

[davidhewitt]

I've always used #[ignore] to disable tests that are very expensive to run, rather than to not run tests that don't work at all. After all, what is the point of cargo test -- --ignored then? So I would personally prefer:

Good point, yes, tests which don't ever expect to work should use #[cfg(not(...))] - my bad 😄

[davidhewitt]

This is really great, thanks again! 💯

[hoodmane]

Thanks for the help and the quick reviews @messense and @davidhewitt! Hopefully I'll get to some followup PRs soon.