[huguesb]

-p is similar to -m except that it will recursively typecheck submodules.

Unfortunately the early module walk was being done with search paths that
did not benefit from site_packages for the selected python executable,
which resulted in some packages being typechecked fine with -m, but
rejected as not found by -p.

Fixes #9386

[gvanrossum]

This looks like it should work, but I'd like a small refactoring to reduce the code duplication.

Also, please add a test. Instructions in test-data/unit/README.md. You could add a test to cmdline.test.

[gvanrossum]

(In the future, please don't amend or squash commits. Code reviewers prefer to see changes resulting from code review separate from the original changes. We'll squash when we land.)

[huguesb]

(In the future, please don't amend or squash commits. Code reviewers prefer to see changes resulting from code review separate from the original changes. We'll squash when we land.)

Sorry, I'm used to code review tools that gracefully handle force pushes and can show diffs between successive updates... I will avoid force pushes here in the future.

[huguesb]

This looks like it should work, but I'd like a small refactoring to reduce the code duplication.

Also, please add a test. Instructions in test-data/unit/README.md. You could add a test to cmdline.test.

The framework for cmdline.test explicitly disables site-packages. I'm modifying the pep561 test suite instead, since it already provisions a typed pkg in a venv.

[gvanrossum]

Thanks for this contribution!