Generating Specialized Rules and Programs for Demand-Driven Analysis

Many complex analysis problems can be most clearly and easily specified as logic rules and queries, where rules specify how given facts can be combined to infer new facts, and queries select facts of interest to the analysis problem at hand. However, it has been extremely challenging to obtain efficient implementations from logic rules and to understand their time and space complexities, especially for on-demand analysis driven by queries. This paper describes a powerful method for generating specialized rules and programs for demand-driven analysis from Datalog rules and queries, and further for providing time and space complexity guarantees. The method combines recursion conversion with specialization of rules and then uses a method for program generation and complexity calculation from rules. We compare carefully with the best prior methods by examining many variants of rules and queries for the same graph reachability problems, and show the application of our method in implementing graph query languages in general.