I have seen magic done with custom types and aggregates. For example, I know of projects with run-length-encoded bitsets and custom aggregates for doing set ope

I have seen magic done with custom types and aggregates. For example, I know of projects with run-length-encoded bitsets and custom aggregates for doing set operations and counts, sort of like Redis' bitset commands but built into PG. This is a massive space optimization, because otherwise you'd just have a join table with zillions of tiny rows.