Eric, that is a good point, but if someone forgets (like I did) or just hasn't gotten around to bumping the number yet, then the build breaks because the interpreter crashes.  I think we should always try to avoid building an interpreter that is in an inconsistent state.

Anyway, I am hitting another problem now -- _freeze_importlib is *not* idempotent.  Thus adding this rule might cause some noise in the builds because importlib.h is different when nothing has actually changed.  I am still investigating that problem.

Assuming I can fix the idempotency problem, then maybe _freeze_importlib should just be always run.