Right. If you don't cache the ordinal, than "next sibling" and "previous sibling" operations have to scan the array. But it's a relatively easy performance enh

Right. If you don't cache the ordinal, than "next sibling" and "previous sibling" operations have to scan the array. But it's a relatively easy performance enhancement to have a validity flag for the ordinals, and only rewrite them when a read requires them. That way, write-heavy workloads won't be punished.