One possible trick for unions is to use bloom filters, providing you have some way of hashing types. If you create a bloom filter `Bloom(t1, t2)`, it's the same

One possible trick for unions is to use bloom filters, providing you have some way of hashing types. If you create a bloom filter `Bloom(t1, t2)`, it's the same as doing `Bloom(t1) | Bloom(t2)`, where `|` is just bitwise-OR. Obviously not perfect as it can produce false positives, but if we keep a filter of sufficient size, this will be low, and still be more space-efficient than keeping a bit per type in a large type hierarchy. Intersections can also be done, but with a potentially higher false-positive rate. The result of `Bloom(t1, t2)` has at least the bits set by `Bloom(t1) & Bloom(t2)`.