These __aenter__ and __aexit__ calls will fail if you modify/extend your test example and annotated the parameter types of these two methods on the Manager class accordingly. None will not generally be a valid type for all of the arguments.

We do not necessarily need to be very strict with the passed argument types here (since we're only using it to add a diagnostic hint, and it seems fine to emit it even if that call would fail when passing the wrong parameters). So I guess it's fine to pass Type::unknown() here instead? Another option would be to allow not just Ok(_) results, but also Err(CallDunderError::CallError(..))?

Hey ! Thank you for the review ! If I understand correctly you want to add the sub-diagnosis event if the code was not annotating the types correctly ?
Example :

class Foo:
    async def __aenter__(self):
        ...
    
    def __aexit__(self,
                 exc_type: str, # <= wrong type
                 exc_value: str,
                 traceback: str
                ):
        ...
        
with Foo():
    ...

But even in this case, we should provide the sub-diagnosis even if the provided typing is incorrect. I guess it makes sense. I am just too unexperienced yet, should I implement it to take care of this case ? But then should we also provide the sub-diagnosis if the user is passing the wrong number of arguments ?

class Foo:
    async def __aenter__(self):
        ...
    
    def __aexit__(self,
                 exc_type,
                 exc_value,
                 traceback,
                 extra_arg
                ):
        ...
        
with Foo():
    ...

But then should we also provide the sub-diagnosis if the user is passing the wrong number of arguments ?

Probably? I think you could achieve this by following this route:

Another option would be to allow not just Ok(_) results, but also Err(CallDunderError::CallError(..))?

I think we should pass Type::unknown() anyway. Passing None is just wrong.

I will address this and add more testing then !

Just to clarify, in case it was unintentional: you're still using Type::none in your latest commit, instead of Type::unknown.

Sorry I just changed it in the last commit. Thank you very much for bearing with me !

This can be simplified to

I think.

Thank you very much for all the reviews you gave, I really learned a lot during this PR about python, rust and ty !