Thanks for the explanation! Sorry for the late reply! I’m not professionally trained in PL and type theories, so I needed to think about this more… Based on my personal experiences with Julia’s package development, I think the current type system has already put a considerable burden on the compiler when doing type inference for complicated composite types (e.g., overspecialization). We should be more cautious when considering any future modifications to the system. I do hope there’s a solution that can be implemented when the time is right.