Handle cwd correctly in pyinfo

[hauntsaninja]

This fixes a recent regression introduced by the change to
use sys.path

Fixes #12956

[hauntsaninja]

cc @AWhetter @ethanhs

[hauntsaninja]

In the issue, Ethan mentioned an alternate possible resolution is to also search the pythonpath for stub packages. I think that's a slightly larger change and we can always do it later. I think we'd want this PR regardless, since it keeps behaviour more consistent across mypy --python-executable ..., python -m mypy, mypy

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[emmatyping]

Thank you! Good catch.

[hauntsaninja]

Just ran into another behaviour that #11143 changed that this PR changes back to that of mypy 0.961.

mypy silences errors on package_path. Since cwd stopped being on package_path, even if we had an editable install of cwd, we stopped silencing these errors. I think the original + this PR, silencing behaviour is actually correct, but I'm not 100% sure.

mypy/mypy/build.py

Line 2458 in a6166b2

for dir in manager.search_paths.package_path + manager.search_paths.typeshed_path:

This wouldn't usually affect mypy, since people rarely check a single module within a package, so we'd have a path here for any BuildSource

mypy/mypy/build.py

Line 1821 in a6166b2

if not path and source is None:

and not go through findmodule.

(It's still a little unclear what I'd need to do to get mypy to report the pandas errors I was seeing via stubtest. Maybe a configuration thing?)