Conversation
|
Note, the docs failure is unrelated to the changes here and being fixed in #10138 |
|
Digging into this failure a bit more as part of the mindeps testing in #10161, it looks like the underlying issue here is that for Based on the testing in #10161, it seems like Zarr roundtrips work as expected with import dask.array as da
import zarr
da.to_zarr(da.ones(3), 'tmp.zarr', overwrite=True)
da.from_zarr(zarr.DirectoryStore('tmp.zarr'))So our options here could either be:
I don't really have a strong preference here, though based on feedback in #10161 seems like we would probably err towards bumping versions? For reference, 2.12.0 is ~9 months old. |
|
Thanks @charlesbluca. Yeah, I have a preference for just setting 2.12.0 as the minimum. Given the current |
charlesbluca
left a comment
There was a problem hiding this comment.
Cool! In that case, how do you feel about reverting the changes to the environment file (since testing should be getting handled in #10161)?
|
Sounds good -- done |
|
Superseded by #10161 |
Offline I saw a user run into the following issue with the latest
daskrelease and an old version ofzarr(in this example I was usingzarr=2.7.1):This is due to
dasknot supportingzarrversions that old. I tried a more recent version (zarr=2.12.0-- released June 23, 2022) and the example worked.I've documented
zarr=2.12.0as our current minimum version in our install docs and pinned to2.12.0in our Python 3.8 build (this is a fine, but low-effort measure -- #10038 would be more complete).@rabernat @joshmoore do you have any recommendations on a better minimum
zarrversion?