This deserves a closer look. It appears to be making use of the ORC format used in Hive, and last I looked was implemented in Java and tied in tight with the Hive arch. So this must be a rewrite of an ORC reader/writer. Selfishly I wish they'd picked Parquet instead of ORC.