Workaround mypy issues with importlib.metadata

[bhrutledge]

Closes #550

There were two errors on the failed build:

twine/__init__.py:24: error: Name 'importlib_metadata' already defined (by an import)
twine/__init__.py:27: error: Module has no attribute "metadata"

The first is an open issue on mypy: python/mypy#1153, and easily resolved with a # type: ignore on the except import. I don't know why we weren't getting that error before.

The second only happens in Python <3.8. I don't know why, since metadata does seem to be an attribute of both modules. Weirdly, removing the conditional import cleared the error.

I opted for the expedient solution of always using importlib_metadata, even though it's redundant in Python 3.8. If folks feel strongly, I can investigate further.

[jaraco]

Sorry, but I'm -1 on this change. It's recommended not to require importlib_metadata on Python 3.8, in part because there are outstanding issues, but also because I've encountered at least half a dozen other packages that are working to avoid using the third-party backport on Python 3.8, so this change runs counter to that intention. I'd started out not having a strong opinion on the matter and slightly preferring importing the backport on 3.8, but there's a clear preference out there for the split behavior.

More importantly, I feel like this workaround sidesteps the issue, degrades the intended behavior, and leaves other projects to run into the same issue. I think I'd rather have a solution that we would recommend to any project using importlib.metadata.

It seems a little bit of a mess that a linter (mypy) has made a very common pattern non-viable. Maybe mypy should support this use-case.

[bhrutledge]

@jaraco No worries. It was interesting to dig into. I made a repository to reproduce the errors and try out some fixes: https://github.com/bhrutledge/mypy-importlib-metadata. That's currently passing, but the commit history has more details. I thought it might be useful for reporting to mypy.

What do you think of this approach?

[bhrutledge]

@jaraco I mentioned this on the related mypy issues, and got a helpful response: python/mypy#1393 (comment). This now uses sys.version_info to determine the proper import.

[jaraco]

This approach is definitely viable. It still feels like this change has implications for the language as a whole, mainly that the Pythonic way of try/except for import is no longer viable. I'll follow up in the upstream issue. Thanks for tackling this issue.