Public s3 prefix without auth

[pgarrison]

Trying again to do #25 but with the tests fixed this time!

Context

Previously bioio could not read from a public s3:// URL without AWS authentication due to this ome-zarr issue. As of 0.9.0 the issue is resolved.

Changes

• Upgrade ome-zarr dependency
• Pass an fsspec instance to ome-zarr instead of just a path
• Add a test with a public s3:// URL

Testing

pdm run pytest bioio_ome_zarr/tests/test_s3_read.py

[pgarrison]

Draft PR until #24 is merged.

[pgarrison]

Previously Dan asked:

LGTM. So if someone wants BioImage(zarr_s3_url), do they really have to do BioImage(zarr_s3_url, anon=True) ?

And I answered:

They must either provide AWS credentials or set anon=True. I agree the usability could be improved because setting anon=True for the same with an https URL will cause an error. I believe this should be the case not just for OME ZARRs but for any file read from a public s3:// path, since I expect all our readers use s3fs under the hood.

So if we wanted to improve the usability I think it happens in bioio or bioio-base by wrapping the BioImage __init__ logic with something like this:

try:
    # __init__ with user's fs_kwargs
except SomethingSpecific as e:
    if protocol == "s3://":
        # __init__ with user's fs_kwargs plus {anon: True}
    else:
        raise e

[BrianWhitneyAI]

Isnt this a dependency of ome-zarr?

[pgarrison]

Yes I believe so, but now we are using it directly

[toloudis]

this is good and valuable documentation but slightly unfortunate that it lives here in the reader's README. It would be great if somewhere in the bioio docs we said something like "be sure to check the readmes or documentation for individual readers, for their specific parameters." It might already be there... This is just an occupational hazard of having readers in separate packages.

[pgarrison]

This particular piece of documentation may apply to all/many readers.

[pgarrison]

Filed an issue for it bioio-devs/bioio#60