Post

now uses O(1) hash table instead of O(N) linear scanning

For all sizes? Linear might still win for small sizes

basically same in testing (except some extra bytes), don't want to have to support 2 different backends