Return type annotation for importlib.metadata.version is incorrect

[DMRobertson]

To my surprise, I discovered in matrix-org/synapse#12223 that it is possible for importlib.metadata.version("foo") to return None. However, the stub suggests version and Distribution.version will always return a string:

typeshed/stdlib/importlib/metadata/__init__.pyi

Line 161 in 50107e6

def version(self) -> str: ...

typeshed/stdlib/importlib/metadata/__init__.pyi

Line 213 in 50107e6

def version(distribution_name: str) -> str: ...

To prove that version sometimes returns None:

1. Create a new virtual environment. I'm using CPython 3.10.2 as my interpreter.
2. Within the venv, pip install bottle. (Any package will do; I choose bottle because it's small and doesn't have any dependencies).
3. The version of bottle agrees with what you've just installed:

   $ python -c "import importlib.metadata as m; print(repr(m.version('bottle')))"
   '0.12.19'

4. Here's the dirty bit: remove the metadata files for that package but keep the metadata directory.
   • Use pip show bottle to find the site-packages location
   • From there, remove all files in the bottle-VERSION-.dist-info directory: `rm /path/to/site-packages/bottle-VERSION.dist-info/*'.
5. The version of bottle is now judged to be None:

   $ python -c "import importlib.metadata as m; print(repr(m.version('bottle')))"
   None

A few other notes:

• For a package foo that's not installed, version("foo") raises a PackageNotFoundError.
• I don't know if this ought to be considered a bug in the library rather than the annotations?
• I haven't dug much into the internals of importlib.metadata but it looks like the version is retrieved by looking inside a metadata attribute on a Distribution. I think other lookups are vulnerable to a similar problem; the only example I could find is Distribution.name.

[DMRobertson]

I'd be happy to PR an update to the annotations making version and name return str | None---but I'm not sure if that's the correct thing to do. Perhaps the implementation ought to be changed to raise an exception in this case, and this ought to be a bug against CPython and the backport importlib_metadata?

[JelleZijlstra]

Not too familiar with importlib.metadata either, but this does sound like it's better treated as a bug in the implementation. The documentation is at https://docs.python.org/3.10/library/importlib.metadata.html#distribution-versions. It's not very precise but doesn't say anything about version() returning None in exotic cases.

[DMRobertson]

The documentation is at https://docs.python.org/3.10/library/importlib.metadata.html#distribution-versions.

There's also the backport of this module, which (strangely?) has a more complete reference. https://importlib-metadata.readthedocs.io/en/latest/api.html#importlib_metadata.version doesn't suggest anything about None there.

this does sound like it's better treated as a bug in the implementation.

Thanks for confirming! I'll get filing issues elsewhere.

[JelleZijlstra]

Thanks! If you're reporting a bug, I'd suggest the docs should also state that it throws PackageNotFoundError for a package that is not installed. :)

[AlexWaygood]

There's also the backport of this module, which (strangely?) has a more complete reference. https://importlib-metadata.readthedocs.io/en/latest/api.html#importlib_metadata.version doesn't suggest anything about None there.

If I understand correctly, I believe changes are usually made to importlib_metadata first, and then "forward-ported" to CPython (don't ask me why). So this isn't that strange :)

[DMRobertson]

For completeness: I've raised https://bugs.python.org/issue47060. It looks like python/importlib_metadata#371 might be related too.