fsspec==2025.3.2.* does not exist

[piskunow]

Dependency fsspec==2025.3.2.* in requirements.txt does not exist.

See https://pypi.org/project/fsspec/2025.3.2/

[martindurant]

$ pip install s3fs==2025.3.2
Successfully installed aiobotocore-2.21.1 aiohappyeyeballs-2.6.1 aiohttp-3.11.15 aioitertools-0.12.0 aiosignal-1.3.2 attrs-25.3.0 botocore-1.37.1 frozenlist-1.5.0 fsspec-2025.3.2 idna-3.10 jmespath-1.0.1 multidict-6.3.0 propcache-0.3.1 python-dateutil-2.9.0.post0 s3fs-2025.3.2 six-1.17.0 urllib3-2.3.0 wrapt-1.17.2 yarl-1.18.3
$ python
python 3.13.2 | packaged by conda-forge | (main, Feb 17 2025, 14:02:48) [Clang 18.1.8 ] on darwin
>>> import s3fs

works fine

[jykae]

@piskunow @martindurant Verified that 2024.3.1 is the latest working version, hope you could fix this

[martindurant]

2024.3.1 is the latest working version

What do yu mean? In my snippet I showed that installing and importing 2025.3.2 works as expected.

[jykae]

@martindurant Installing exact version seems to work fine, you're right, but resolving for example with pipenv does not seem to work if specify "==2025.*" for example to Pipfile

[martindurant]

"s3fs~=2025.3" is probably what you want? pip's version spec does not allow for a single digit ("2025") as the version, and this has nothing to do with s3fs's requirements.

For your specific invocation: yes, it does work!

$ pip install "s3fs==2025.*"
Successfully installed fsspec-2025.3.2 s3fs-2025.3.2

I would very much assume that pipfile does the same as pip, but if not, that is their problem.

[jykae]

Tried same ==2025.* specifier with with uv and works, looks like issue with my current pipenv. Maybe should start moving to uv.

[piskunow]

Hi all, apologies I couldn't add more details earlier.

I found the issue when installing a library from JFrog Artifactory, which has fsspec as a dependency on a Databricks Cluster with 15.4 runtime.

I could not reproduce the issue locally with either pip or poetry install (these worked OK).

Neverthelss, the specification fsspec==2025.3.2.* is the only dependency that caused problems, and the only one with the extra asterisk after the thrid dot (.) .

My guess is that version resolution is not correctly done at some point: either the python package manager on Databricks side, or the listing of versions with this glob pattern on JFrog Artifactory.

I could followup on the issue with JFrog or Databricks, but I would also suggest to use a common pattern instead of fsspec==2025.3.2.* here.

[martindurant]

After a decent amount of thinking, I will allow this - it is, after all, only going back to exactly what we had before.