I wonder if there is a memory consumption tradeoff for this new data structure? Based on a few initial implementations I see in github, looks like it may be significant? Still a nice discovery.