This (implicitly) changes the type signature of issubdtype to (arg1: object, arg2: object) -> bool. And I'm sure you see why I (a static typing nerd) don't really like that haha.

But on a more serious note, this will cause np.issubdtype(np.int8 | np.uint8, np.integer) to return False. If we ever decide to implement support for such types.UnionType input (e.g. by simply using typing.get_args()), then that the result would change to True. But that is backwards incompatible. So the consequence of this PR, is that we are no longer able to implement support for UnionType, or any other extensions to the allowed input types. So this PR is effectively puts issubdtype into feature-permafrost. This also means that if we ever were to add a new dtype or scalar type, it would be backwards- and forwards-incompatible for issubdtype.

@jorenham

Thank you for your comment! Your concerns make perfect sense.

As you pointed out, this change breaks the type signature and removes the possibility of future extensions.
I'm planning to close this PR once @seberg also agrees.

In any case, I really appreciate your detailed feedback. I learned a lot from your review. Thanks!

Yeah, I agree, not much point in doing this and it may be confusing.