Errata for "Building Efficient Query Engines in a High-Level Language" (PVLDB 7(10): 853-864)

This is in response to recent feedback from our peers that calls for a number of clarifications regarding the experimental section of our paper. Table 1 clarifies which optimizations are used in each evaluated flavor of LegoBase. HyPer-simulated is a configuration of the LegoBase codebase that mirrors the HyPer system as closely as possible by just activating some of the main optimizations and deactivating others: using the push engine, with operator inlining active but data structure optimizations and data layout transformations turned off. In addition, as noted in our paper (footnote 8), the actual HyPer system uses query plans generated by its own query optimizer while HyPer-simulated uses query plans from DBX. We use TPC-H queries and generated data at scaling factor 8 to evaluate the impact of our compilation techniques. For all evaluated systems, reported query evaluation times only include the execution time of the query and exclude the time taken for query optimization/compilation and loading the data into the main-memory data structures. We note that the data structure specialization optimizations of Section 3 of our paper (whenever used) are a gentle form of pre-computation, as would arguably be loading row data from disk into a main-memory column store.