I solved that with the brute-force method of just reading the pg_class tuple an extra time. I think this is probably good enough, really. I thought about having do_autovacuum pass down the tuple to table_recheck_autovac so as to not increase the net number of syscache fetches, but I'...