Describe the issue:

Hi folks, your dask._compatibility.import_optional_dependency seems not to work correctly, since even if there is absolutely no Xarray installed in the current environment, that utility still returns some type of strings, definitely not None nor False as it should; consequently, Dask thinks Xarray is there and is trying to access modules like xr.DataArray, which invariably fails.

Minimal Complete Verifiable Example:

from dask._compatibility import import_optional_dependency

xr = import_optional_dependency("xarray", errors="ignore")
print(xr)

that returns a dummy module <module 'xarray' (namespace) from ['/home/valeriu/xarray']> in my case, which is enough for Dask to be happy xarray was successfully imported. And yes, I do have an xarray dir in my $HOME, which is Xarray's source code cloned from GitHub, but it's not installed in any way.

Anything else we need to know?:

This affects higher level tools like Scitools iris SciTools/iris#6723 or anyone else that doesn't have Xarray as direct dependency from their package.

Environment:

• Dask version: 2025.09.1 - conda-forge
• Python version: 3.13.7
• Operating System: Ubuntu 24.04.3 LTS
• Install method (conda, pip, source): conda-forge

Cheers 🍻