Also, while PyThreadState_GetDict() is the inspiration here, we don't have to copy it exactly.  For instance, PyInterpreterState_GetDict() takes a PyInterpreterState* argument, whereas PyThreadState_GetDict() takes no arguments and gets the PyThreadState* from thread-local storage.

Is there anything else that would make sense to do differently with PyInterpreterState_GetDict()?  It's pretty basic, so I'm guessing "no". :)