Sebastian Rittau

This needs a bit more infrastructure work in typeshed. Currently, our tests don't install the dependencies from `METADATA.toml`, meaning that the tests could fail due to changes in cryptography. I...

There's been no release since. We can remove this six months after a release with `py.typed`.

> Temporarily pin to 3.10.5 for all our stubtest runs? I would say that's the best option.

I agree with @JelleZijlstra. Let's do whatever the stubbed library does. I'm absolutely in favor of removing `__nonzero__` in the libraries themselves, though, if they dropped Python 2 support.

The xml stubs are indeed incomplete. The marker for a missing return type is not `...`, though, but the missing `->` operator. For example: ```python def createDocumentFragment(self): ... # no...

I'm fine with changing `exclude` and `filter`. The approach I find most intuitive was to just run `mypy_test stdlib/foo.pyi` or `mypy_test stubs/bar` to test a specific file or directory. `--verbose`...

I know this is quite controversial, but I would like to see this go in rather sooner than later. The longer we wait the worse it gets, and our stubs...

> I prefer `Iterator[T]` because it's simple. But it's also confusing. A generator function return a generator after all (which happens to be an iterator as well).

I'm still in favor of allowing `Generator[X]` for cases like this. `Iterator[X]` is just the wrong annotation and as this discussion shows problematic in some cases.

I am opposed to merging this. If there isn't a fix for mypy until the next release, I suggest that mypy ships a patched version of typeshed.