Julia Signell

Yeah that test wasn't passing on a build that we don't do anymore (the windows32 python3.7 build) so hard to say if it's not passing or not anymore.

Yes we probably should not have closed this just because we stopped testing with the conditions that created it :)

Yeah that seems like a super reasonable solution @detrout. I haven't really heard of people experiencing issues with this in practice. Seems like the issues are more around the tests...

Please feel free to open a pull request with your suggestion!

add to allowlist

> I believe the failing tests are because the functionality for those cases has changed since the shape of the array is not unknown anymore. That's an exciting way to...

I think you should look through the failing tests and set `arr._chunks` explicitly. Something like we do here: https://github.com/dask/dask/blob/377940b64914111ff92f2ea7f0b264dccabe273a/dask/array/tests/test_array_core.py#L3800 I was originally trying to suggest that you could make a...

I think some of the tests might be failing because the number of output chunks is different when using a boolean filter than it was before. Before the number of...

> One of these days, I think we should reconsider some of the policies around all of these backwards incompatibility checks, and probably phase some of them out. This has...

I would think the easiest solution is to raise an error as high up as possible. Without known divisions we can't be sure that the data is sorted by the...