[red-knot] Find conflicting declarations for attribute in class methods

[smokyabdulrahman]

Summary

Handling one more case for class attributes: checking for conflicting declarations for class attributes in methods.
part of #15964

Test Plan

No test for now

[MichaReiser]

I put this back in draft as it doesn't seem finished.

[sharkdp]

I put this back in draft as it doesn't seem finished.

Just for additional context, I suggested splitting the implementation into two parts: (1) collect all conflicting declared types so we would in principle be able to emit a diagnostic — assuming we could do it from anywhere, and (2) actually create a new diagnostic and make the required changes to be able to emit it.

But even part (1) doesn't seem finished, so keeping this in draft seems fine. @smokyabdulrahman Let us know if you need additional help to proceed here.

[smokyabdulrahman]

I put this back in draft as it doesn't seem finished.

Just for additional context, I suggested splitting the implementation into two parts: (1) collect all conflicting declared types so we would in principle be able to emit a diagnostic — assuming we could do it from anywhere, and (2) actually create a new diagnostic and make the required changes to be able to emit it.

But even part (1) doesn't seem finished, so keeping this in draft seems fine. @smokyabdulrahman Let us know if you need additional help to proceed here.

Hey @sharkdp,
I couldn't think of any other place to tackle and add the logic of collecting conflicting declared types to.
Also, I am not sure how to test for part (1) even in a hacky way.

[sharkdp]

I couldn't think of any other place to tackle and add the logic of collecting conflicting declared types to.

We identified two spots in the linked issue. If I understand it correctly, you solved it for one of these, but the more difficult one would be in the change to the loop logic. I can go into more details tomorrow, if that would help.

Also, I am not sure how to test for part (1) even in a hacky way.

The infrastructure for emitting diagnostics from Type::member (and friends) is still not in place. If you're okay with waiting for that, you could still test your implementation by running Red Knot manually on a test file, and "emitting" diagnostics via eprintln! or similar.

The important thing for part 1 would be to get the semantics correct, i.e. to identify two or three test cases that would clearly show that we would be able to generate the right diagnostic messages.

If that sounds too complicated, I'm also okay with leaving that open for a bit longer, and revisit it once the diagnostics infrastructure is in place.

[sharkdp]

The infrastructure for emitting diagnostics from Type::member (and friends) is still not in place

This is being tracked in #16298