Hi, thanks for the feedback. Please take a look at the new revision of this proposal, which abandons reference counting entirely. (Please also reply on the new thread.)
That was the original proposal. It was not received well; making PyInterpreterState * implicitly “strong” or “weak” was very confusing.