Fix incorrect name lookup for decorated methods

[TH3CHARLie]

Resolves #8161

According to comments of lookup in mypy/semanal.py, when we look up a class attribute, we require that it is defined textually before the reference statement, thus line number is used for comparison.

When function has decorators, its line number is determined by the top decorator instead of the def. That's why #8161's code fails because on line 8, the A in Type[A] has the line number of 8 while the @staticmethod function A has the line number of 7 due to the decorator.

Thus we need to properly handle this by introducing the number of decorators when deciding textural precedence.

[ilevkivskyi]

Thanks! Generally looks good, I have few minor comments.

[TH3CHARLie]

oops, I made some stupid mistakes

And seems like the @overload case does not apply to this fix

[TH3CHARLie]

I add new logic in semanal.py to check if the class attribute we are accessing is an overloaded function if so, we check whether we are referencing it from a function with the same name, i.e. a variant of these overloads

[ilevkivskyi]

Thanks for updates! Here are couple more comments.

[ilevkivskyi]

This is still not ready, I have few more comments.

[TH3CHARLie]

Good catch and thanks for the careful review! I check the logic again and indeed there are some mistakes, here's the update:

• update logic in is_textually_before_statement to ensure if we are referencing an overloaded item, the function return False
• add negative test
• For the logic in is_overloaded_item, I have to point out that neither my previous solution nor the solution purposed by you work without modification.
  In this case, we need to check if a FuncDef is one of the overloaded variants and these variants are expressed in either Decorator or FuncDef. The former one is tricky as we need one more step to extract its decorated function for checking. I think we can safely ignore the case when statement i s a Decorator here

[ilevkivskyi]

Almost ready now, just couple small comments.

[TH3CHARLie]

To address that node.impl can also be a Decorator, the logic would be long enough to make it hard to understand, so I deliberately separate single return into two booleans to make the code easy to understand.

[ilevkivskyi]

LGTM! (assuming the tests pass)