I am not typing expert but I think we need TypeVar(T, bound='array') now (PEP 484), and in the future Self (PEP 673). The current approach would not work for subclasses. My vote would go for the status quo (no type hint for self) but let's wait for others to chime in.

array is what the other methods use so keeping that here is fine. We can open another issue about using a different type hint for self.

CI failure is unrelated, let's retry that.

Sorry, so this PR does two things:

1. It fixes the missing self to the array methods (which I overlooked)
2. It adds a type hint for self

We can open another issue about using a different type hint for self.

Let's do that.

Okay, CI is unrelated, I'll open a separate issue for that too.

Yeah the CI failure is likely a permission issue that we also hit before.

The type hint for self is one instance of a general pattern; same as for scalars we use int and not anything that ducktypes with int (which can be type-able by SupportsInt or some such thing).