Compiling JSONiq Queries to Native Machine Code: Semester Project

Abstract

JSONiq is a querying language specifically tailored for JSON files, whose key capability is to deal with unorganized and heterogeneous data. However, this benefit comes at a computational cost. The current implementations usually do not have any optimization on homogeneous data. In some cases, these optimization exists but are however limited. As a result, a substantial amount of overhead can be removed when considering homgenous data. We introduce our work for this exact purpose. Our proposal extends past work around the MLIR framework by compiling directly into native machine code. Our findings suggest that, for a selection of queries, our approach surpasses the current implementations with speed-up of around 40.