Encourage use of `importlib.metadata.version`

[jakkdl]

Description

I only recently found out about importlib.metadata.version as a way of getting the installed version of a package; as opposed to reading some global __version__ variable that may or may not be properly exported.
I would love a rule that encouraged the use of the more portable & reliable importlib variant, although I have a feeling that a simple rule that detected any from ... import __version__ would raise a bunch of false alarms on packages that export it as ways of versioning ... other things than the package itself. (maybe only trigger if importing from the top namespace of a module?) Or idk other weird stuff.

Anyway, thought I'd throw out the idea anyway and see if it sticks.

[AlexWaygood]

This might be controversial for authors of CLIs, since importlib.metadata has historically had quite a high import time. Importing importlib.metadata and using importlib.metadata.version() could mean that running <your_cli_app> --version takes longer if it uses importlib.metadata than it would if it just checked __version__

[jakkdl]

This might be controversial for authors of CLIs, since importlib.metadata has historically had quite a high import time. Importing importlib.metadata and using importlib.metadata.version() could mean that running <your_cli_app> --version takes longer if it uses importlib.metadata than it would if it just checked __version__

I was thinking of it more when interfacing with 3rd-party packages rather than within-package use; so you you'd probably want to exclude those imports.

But it definitely is slower, I did some quick benchmarks with

import six
import importlib.metadata
print(importlib.metadata.version('six')

vs

import six
print(six.__version__)

and the former averaged .0642s and the latter .0174 for a ~4x difference (py3.13; on py3.9 I got .049 and .022)