At first quick glance, this makes sense and the API looks reasonable.

Question: what happens on interpreter shutdown?

Shutdown obviously finalized and clears out most all dicts.  I guess the C callback simply gets called for each of these?  That makes sense.  Just wondering if there are any ramifications of that.  The callback is in C so it shouldn't have issues with this.

A pyperformance suite run on an interpreter modified to support this but having no callbacks registered would be useful.  (basically judging if there is measurable overhead added by the watched bit check - I doubt it'll be noticeable in most code)