Fix bug when TypeOf is one of options in OneOf / AllOf

[MapleCCC]

This PR should fix #528.

I also added unit test, to prevent regression.

[MapleCCC]

The reason I choose to disallow combination of AllOf and TypeOf, is because I think this is analogous to how the combination of AllOf and OneOf is disallowed. It seems natural to think of TypeOf and OneOf having similar semantic, although on different levels. Please feel free to let me know if you prefer other design.

[MapleCCC]

Static type checkers didn't catch this bug for us. I wonder if we can do something in our code to help type checkers catch these bugs for us.

The logic in the matchers sub-system are large and complicated, making it hard for human developers to manually detect all kinds of logic bugs by eyes. It would be a huge gain if we manage to have type checkers help us here.

[zsol]

Thanks! I agree with your sentiment, but don't see a way to express these kinds of constraints using Python's type system atm. Tests are our only tool here unfortunately

[MapleCCC]

Thank you for the comment.

I am more optimistic on this though. I have a hunch that we can greatly improve type-check-ability here if we designate some of the types in libcst.matchers package to be so-called sealed type [1].

This feature is actively discussed in Python typing community recently [2]. Current workaround seems to be to use union type (a.k.a. sum type) to simulate sealness.

I will play with this idea a bit more, and try to see whether it brings outweighing benefits here.