downcast regression: ambiguity due to global where-bounds

[compiler-errors]

https://crater-reports.s3.amazonaws.com/pr-133502-1/try%23fa8e241660363f48d64b66b05eea58c93ab828fb/reg/mockall-0.13.1/log.txt

pub trait Any: 'static {}

pub trait Downcast<T>: Any
where
    T: Any,
{
}

impl dyn Any {
    pub fn is<T>(&self)
    where
        T: Any,
        Self: Downcast<T>,
    {
    }
}

[compiler-errors]

I believe this has to do with a disagreement in region responses between the param-env bound we get from elaborating Self: Downcast<T> -> Self: Any (where Self = dyn Any + 'static) and what we get from the built-in bound dyn Any: + 'static: Any.

This specifically requires that 'static as a super bound, suggesting that an extra '?0: 'static that comes from confirming the built-in object impl's super bounds is preventing us from merging two trait candidates.

[compiler-errors]

Worst case we wait until we get better w/ regions in responses 💀

Best case we come up with some better fix to the way we combine candidates when their region responses are practically equal. (At least, I think that's the case here).