Dependency checking with `importlib.metadata`

[davidhassell]

#361 got rid of dependency version checking in init.py.

However, it could still be done cheaply with the importlib.metadata module (as pointed by @sadielbartholomew: #362 (comment)):

>>> from importlib import metadata
>>> metadata.version("cftime")
'1.6.4'
>>> metadata.version("netCDF4")
'1.7.2'
>>> metadata.version("dask")
'2025.7.0'

So, do we want to do this?

Related, is making sure that our libraries return nice things in this manner.

[sadielbartholomew]

I've briefly investigated and our CF Data Tools are all sufficiently packaged according to standards (maybe not the most modern way to do it, via setup.py, but if/until that's deprecated that's no issue) that the metadata calls work fine, with the note that for cf-python you must use the full distribution name and not the import name:

>>> from importlib import metadata
>>> for lib in ["cfdm", "cf-python", "cfunits", "cf-plot"]:
...     metadata.version(lib)
...     
'1.12.3.1'
'3.18.0'
'3.3.7'
'3.4.0'
>>> # But note 'cf' doesn't work!
>>> metadata.version("cf")
Traceback (most recent call last):
...
  File "/home/slb93/miniconda3/envs/cf-env-313-numpy2/lib/python3.13/importlib/metadata/__init__.py", line 409, in from_name
    raise PackageNotFoundError(name)
importlib.metadata.PackageNotFoundError: No package metadata was found for cf

Apparently, according to <AI tool of choice> [with health warning], allowed mismatch between the distribution name (used by tools like pip and importlib.metadata) and the import name is a common pain point in Python packaging. When I asked how we might be able to handle the unideal situation of metadata.version("cf") not working when that's our import name that folks would be most likely to try (at least, first), it suggests options of:

• wrapping to ensure we have a cf.__version__ that works - we already do this so, done;
• change the distribution name to match the import (cf) i.e. setup(name="cf", ... but we can't do that due to cf already being a PyPI package - namespace taken;
• provide a lightweight “alias” package: "keep cf-python as the canonical distribution, but publish a second package cf that simply depends on it and re-exports it." - but that would be silly.

So, at least from that response plus a little search engine explore on key terms, there isn't all that much we can do about that, unless you have any ideas?

[sadielbartholomew]

(Updated the above comment to correct that the metadata.version does work for cf-plot, I just didn't have it installed in me new-ish Python 3.13 conda env so it failed for that reason, not due to insufficient packaging as first assumed.)