Provenance, Incremental Evaluation, and Debugging in Datalog

The Datalog programming language has recently found increasing traction in research and industry. Driven by its clean declarative semantics, along with its conciseness and ease of use, Datalog has been adopted for a wide range of important applications, such as program analysis, graph problems, and networking. To enable this adoption, modern Datalog engines have implemented advanced language features and high-performance evaluation of Datalog programs. Unfortunately, critical infrastructure and tooling to support Datalog users and developers are still missing. For example, there are only limited tools addressing the crucial debugging problem, where developers can spend up to 30% of their time finding and fixing bugs. This thesis addresses Datalog’s tooling gaps, with the ultimate goal of improving the productivity of Datalog programmers. The first contribution is centered around the critical problem of debugging: we develop a new debugging approach that explains the execution steps taken to produce a faulty output. Crucially, our debugging method can be applied for large-scale applications without substantially sacrificing performance. The second contribution addresses the problem of incremental evaluation, which is necessary when program inputs change slightly, and results need to be recomputed. Incremental evaluation allows this recomputation to happen more efficiently, without discarding the previous results and recomputing from scratch. Finally, the last contribution provides a new incremental debugging approach that identifies the root causes of faulty outputs that occur after an incremental evaluation. Incremental debugging focuses on the relationship between input and output and can provide debugging suggestions to amend the inputs so that faults no longer occur. These techniques, in combination, form a corpus of critical infrastructure and tooling developments for Datalog, allowing developers and users to use Datalog more productively