Spec for sys.meta_path too strict

[jaraco]

In this comment, we learned that typeshed is too constrained on its spec for sys.meta_path, expecting instances of MetaPathFinder, whereas the existing values are of Type[MetaPathFinder], but more importantly, the objects there need not be instances of any particular class, but must conform to a protocol (having a .find_spec() or .find_module() on Python 3 and having a .find_module() on Python 2).

[pfmoore]

I've just hit this as well, here. Is there a workaround until it's fixed?

[layday]

#6344

[Akuli]

I believe #5349 fixed this already, and mypy just uses an old version of typeshed.

[pfmoore]

I believe #5349 fixed this already

Nope, see my review comment on #6344. The current code doesn't allow for class or static methods.

[pfmoore]

Is there any progress on this? I'm reviewing whether to do anything to add type annotations to editables, and this is a blocker for me.

[layday]

It looks like find_module and invalidate_caches were re-introduced in the typeshed at some point.

[layday]

Actually, I was looking at the wrong place (at importlib.abc.MetaPathFinder), and that hasn't changed. Can you explain what problem is it that you are experiencing?

[pfmoore]

Sorry, I can't. My comment that I mentioned above seems to have been lost somehow.

I guess it'll have to wait until I get the motivation to try to apply types again, and I encounter the issue once more...

[abravalheri]

@jaraco would #11890 be enough to address this issue? (it does not handle 2.7, or find_module as it is deprecated).

[jaraco]

Yes, that sounds like a proper fix. Thank you!