Comment 16 for bug 598666

Oh yeah, of course. Indeed, good work Siegfried.
While reading the last few comments one possible fix came into my mind:
Let's maintain a (temporary) helper table called '_fix_cache' with (table_name VARCHAR, id INTEGER) and create a 'BEFORE DELETE' trigger on each cached table (interpretation, manifestation, mimetype, ...) which inserts the table name and the deleted id in _fix_cache once a row in the related table got deleted. All we now have to do is run a 'SELECT * FROM _fix_cache' after deleting the events. If some rows in manifestation/interpretation/... got deleted while deleting events we get back the the ids of the cached values which we then can delete from the caches.
I know that's not optimal, but looks like the only solution to me. This should not be a *big* performance issue, since the number of rows in the helper table should be very low. I'll hack on a prototype tomorrow, and see how it works.