indexing

precursor: b+tree

phext has nice scaling properties for multi-core machines. once you've parsed all of the top-level libraries, for instance, you could envision handing work off to worker threads. this process could be repeated for each dimension break - forking threads early and often in the parsing process.