> In a way it's no different than caching a pointer to an arena-allocated chunk > that may outlive the arena context. Maybe I'll think of migrating resources > up the context tree if their owner changes?
It is not different indeed. And once you have "migration" you effectively have a system with a `wasMoved` state and uniqueness and when you expand that you get reference counting. Which is better left to the programming language as the PL knows the types and the OS doesn't. I did many experiments with bump pointer allocators and regions before deciding on ARC/ORC. Everything else simply does not work as well, all things considered (ease of use, resource consumption, composability).
