This will allow us to have much better Unicode support. It does incur a memory usage regression which we'll have to optimize to cover.