🐛 cluster.hierarchy: use generic return type for DisjointSet.__getitem__

[lispandfound]

Addresses #991

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[jorenham]

Thanks for this!

Could you also add a quick little type-test for this in https://github.com/scipy/scipy-stubs/tree/master/tests , to confirm that it's now working as intended? It would also help me understand it a bit better haha.

[lispandfound]

Yes I will do this, and I will also address the incorrect type in __init__ you noted in the issue.

[lispandfound]

...and an issue with __contains__ accepting object when it should accept _T

[lispandfound]

I made a small change to the signature to leave _T unbound in the case where no elements are passed:

_T = TypeVar("_T", bound=op.CanHash)

But I now see that the test matrix fails for numpy<2.2, is this expected? If so, I will revert that change.

[jorenham]

I made a small change to the signature to leave _T unbound in the case where no elements are passed:

_T = TypeVar("_T", bound=op.CanHash)

But I now see that the test matrix fails for numpy<2.2, is this expected? If so, I will revert that change.

Removing that default would break backwards compatibility. Using default=Any instead of object is what I'd do here.

[jorenham]

But I now see that the test matrix fails for numpy<2.2, is this expected?

The problem is that before NumPy 2.1, there was an issue in NumPy's bundled stubs where the __hash__ methods was missing for certain scalar types like int32. That mistake was easy to make though, because the __hash__ method "disappears" if you only provide an __eq__ method, without also providing an explicit __hash__ method. Pyright, unlike mypy, also applies these cpython runtime semantics to the stubs. So that's why pyright is selectively complaining here.

It's easy to work around though, i.e. set _T's bound to something like bound=op.CanHash | np.generic.

[lispandfound]

Ok I think this is now in a reviewable state again.

[jorenham]

This is looking very good; thanks Jake!