IDE Dataflow Analysis in the Presence of Large Object-Oriented Libraries

A key scalability challenge for interprocedural dataflow analysis comes from large libraries. Our work addresses this challenge for the general category of interprocedural distributive environment (IDE) dataflow problems. Using pre-computed library summary information, the proposed approach reduces significantly the cost of whole-program IDE analyses without any loss of precision. We define an approach for library summary generation by using a graph representation of dataflow summary functions, and by abstracting away redundant dataflow facts that are internal to the library. Our approach also handles object-oriented features, by employing an IDE type analysis as well as special handling of polymorphic library call sites whose target methods depend on the future (unknown) client code. Experimental results show that dramatic cost savings can be achieved with the help of these techniques.