Re: numpy type hints

I had avoided going there previously, partly because we mostly use ArrayLike, which is not subscriptable, and partly because I just didn't want to broach it yet.

I think starting with C-level interfaces for further specifying is reasonable, so I don't have a problem with adding it here.

Just to be sure, these are specifically ndarrays that must be passed, not any array like (lists of floats, etc.)?

The other part that is confusing about npt.NDArray is that it at least has space carved out to specify shape as well, though I don't think that is actually used... so while even in their examples they use the form like you have here, where there is a single parameter, which is the dtype, if you look at the repr of the object it is actually NDArray[Any, <dtype>].

In any case, I think what you have is likely correct, just was always one of the things that gave me a slight pause in trying to use the more specific numpy types.

I had avoided going there previously, partly because we mostly use ArrayLike, which is not subscriptable, and partly because I just didn't want to broach it yet.

I think starting with C-level interfaces for further specifying is reasonable, so I don't have a problem with adding it here.

Just to be sure, these are specifically ndarrays that must be passed, not any array like (lists of floats, etc.)?

Yes, these are all used internally only, and are called with normalized arrays, except for empty ones which are (), which I should correct in the type hints. Also, I kind of assumed that py::array would not coerce, but it does, so the tests also use lists, but I think it would be more consistent to fix those.

The other part that is confusing about npt.NDArray is that it at least has space carved out to specify shape as well, though I don't think that is actually used... so while even in their examples they use the form like you have here, where there is a single parameter, which is the dtype, if you look at the repr of the object it is actually NDArray[Any, <dtype>].

NDArray doesn't have a shape, but ndarray does. However, as far as I can tell, it's not used for anything yet.

In any case, I think what you have is likely correct, just was always one of the things that gave me a slight pause in trying to use the more specific numpy types.