feat: allow classmethods to receive Py<PyType>

[wyfo]

This is necessary for async classmethods

[adamreichold]

Could you add a test case and change log entry for this? Thanks!

[davidhewitt]

Also, if you're willing, it would be great to also support this for #[pyfunction(pass_module)] please. I think that will have the same problem.

[wyfo]

Also, if you're willing, it would be great to also support this for #[pyfunction(pass_module)] please. I think that will have the same problem.

Indeed, I've added it. I had to modify type_is_pymodule function, but I am wondering if this function is really useful, as there is no check like this for #[classmethod].

[codspeed-hq]

CodSpeed Performance Report

Merging #3587 will improve performances by 22.03%

_{Comparing wyfo:classmethod_into (744de3a) with main (abe518d)}

Summary

⚡ 2 improvements
✅ 76 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	wyfo:classmethod_into	Change
⚡	list_via_downcast	153.9 ns	126.1 ns	+22.03%
⚡	not_a_list_via_downcast	153.9 ns	126.1 ns	+22.03%

[davidhewitt]

👍

@wyfo similar finishing request here - can you please squash it? Also, if you're willing to check the patch diff on codecov and add an additional test or two to hit the coverage target, that would be massively appreciated 🙏

[wyfo]

Actually, the issue with coverage comes from the error cases pyfunction::type_is_pymodule, because I've added 4 of them. But they cannot be covered, because test would not compile; actually, I thought that UI test would be counted in coverage, but it seems not, so I've no way to add coverage for these branches. Have I missed something here?
By the way, I've proposed to remove pyfunction::type_is_pymodule, because it's redundant with the type checking, and not consistent with #[classmethod] which doesn't have a macro-level check. It would solve the coverage issue.

[davidhewitt]

Ah makes sense, let's merge this and I'll have a play with what the error message looks like with that check removed 👍