Array-like type support

[kevinsantana11]

Original Thread

Hi @kevinsantana11 / @selipot, I tried to fix mypy for all modules and failed (to say the least). I did this PR as a first step, we have a few arguments that were:

x: int | None = 0

which should simply be:

x: int = 0

To make typing a bit more easy for us, I think we should discuss a strategy. As of right now, we have a lot of functions that try to accept all types:

list | np.ndarray | pd.Series | xr.DataArray

but this is not constant across clouddrift, so for example, another function will have:

np.ndarray[float] | pd.Series | xr.DataArray

so if we pass a list somewhere mypy will complain.

Similarly, we are accepting list at a few places alongside other array_like structures and do operations like (if x is array_like):

x / CONSTANT
x.shape
x.size

which are not valid if isinstance(x, list).

My main questions are:

* do we really want to support all the types?

* what do other libraries do in that situation? Or someone knows what's the best practices here?

One solution could be to defined a clouddrift_array_like_type once, and then reused it.. Something we do in pairs.py with:

array_like = Union[list, np.ndarray, pd.Series, xr.DataArray]

Maybe we can copy this to a discussion?

---

Main Question

• do we really want to support all the types?
• what do other libraries do in that situation? Or someone knows what's the best practices here?

[kevinsantana11]

I need to dig into how other libraries handle this but I think its a good idea to standardize on the types we support.

We should also consider supporting DaskArrays since it can help enable parallelization throughout the library but am not sure on supporting vanilla python lists.

[selipot]

I rarely, if ever, use python lists to represent vectors of data so i think it would be ok for me not to include those in the ArrayLike type.