Consolidate array Dispatch objects#7505
Conversation
|
Should we extend codeowners here as well? cc @pentschev @quasiben (for awareness) |
I think I have a lableling solution that will actually work better given the restriction described here: #7503 (comment) |
jsignell
left a comment
There was a problem hiding this comment.
I think I prefer to have the dispatch objects all in one file separate form any of the registrations and with no imports. That way the registrations can import them all from dispatch and then the registrations don't get imported anywhere, so it makes for clean dependencies. And we can safely import from dispatch in all the places where we explicitly use the dispatch objects.
|
We should hold off merging either of these until we get a chance to talk things through thoroughly with the dask-cudf folks. |
|
Yep, agreed let's hold off for right now. FWIW I think the changes here are backwards compatible. |
|
Looks fine to me, also ran CuPy tests locally and they pass as well! |
|
Just noting this is still of interest, know everyone is busy with the Summit this week, but it would be good to figure out next steps here (maybe next week?). In particular Julia mentioned this PR might need some updating ( #7503 (comment) ). |
|
Yeah good ping @jakirkham. The remaining work is just to make the file structure in this PR match the other one, so separate out backends and dispatch into their own files. |
|
Yeah, thanks for the ping @jakirkham. I just pushed a commit which updates the file structure here to match what @jsignell did over in #7503 |
|
Thanks for testing @pentschev and reviewing @jakirkham @jsignell |
Requires PR ( rapidsai/cudf#8342 ) As these functions recently got relocated, handle `import`ing from the new location with a fallback to the old location. xref: dask/dask#7503 xref: dask/dask#7505 Authors: - https://github.com/jakirkham Approvers: - GALI PREM SAGAR (https://github.com/galipremsagar) - Benjamin Zaitlen (https://github.com/quasiben) URL: #623
This PR consolidates
Dispatchobjects used throughoutdask.array, and their corresponding registrations, intodask/array/backends.py. This goes along with #7503, which does the same thing fordask.dataframe.xref #7502
cc @jsignell @jakirkham @kkraus14