Thanks for working on this! Overall the implementation seems good. I've left some comments on things to address.

Regarding the approach to providing a blocking "send()"... After having seen the block-in-the-function approach here, I still have reservations about it. There's a lot of value to keeping the "send" part separate from the "wait" part.

Here are some alternatives that offer that separation:

• return an acquired threading.Lock which the receiving interpreter releases
• return a wait(timeout=None) function which the receiving interpreter will un-block
• caller passes in an acquired lock that the receiving interpreter releases
• caller passes in a callback function that the receiving interpreter triggers (via _Py_AddPendingCall())

My preference is that first one (return a lock). Since we are crossing the interpreter boundary we need to be extra careful. A user-defined callback is too risky. Of the remaining options the first one is simplest and most consistent conceptually.

FWIW, I expect that most of the code in this PR should still work mostly as-is.