Use asarray in extract

[jakirkham]

Force both array inputs to Dask Arrays. This avoids potentially converting a different array type to a NumPy array by accident.

• Tests added / passed
• Passes flake8 dask

[mrocklin]

This looks fine to me. I wonder how we should end up testing changes like this.

Something with sparse perhaps?

(not a requirement for this PR necessarily, just wondering what the right approach is long term)

[mrocklin]

Also cc @pentschev , just so that he's aware

[jakirkham]

It's a fair question. Though it's probably one that we want to ask over the test suite as a whole as opposed to each test case. Pytest's fixtures come to mind, but IDK if that is the right solution. Should we move this to an issue to discuss further?

[mrocklin]

Sure, that makes sense

…

On Tue, Mar 5, 2019 at 7:26 PM jakirkham ***@***.***> wrote: It's a fair question. Though it's probably one that we want to ask over the test suite as a whole as opposed to each test case. Pytest's fixtures come to mind, but IDK if that is the right solution. Should we move this to an issue to discuss further? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#4549 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AASszPHR6QJYLVp5o-A-gGBN9vqsgJvvks5vTxkygaJpZM4bfz32> .

[pentschev]

I think it would be good indeed to have more tests with Sparse too, in particular for these issues related to NEP-18. I'll try to add some more within the context of #4486.