Using data flow analysis to infer type information in Java bytecode

Abstract

The paper shows how dataflow analysis can be used to infer type information related to Java bytecode, and evaluates this approach. All the elements needed to specify the dataflow problem to be solved are conveniently defined, according to the theory on data flow frameworks, along with algorithms to build them automatically from the Java bytecode. To evaluate the approach, we consider the specific problem of statically determining the set of methods potentially invoked by each method call instruction, which is useful for several purposes, such as building accurate call graphs. Experimental results are presented, showing that data flow analysis can improve the accuracy of the results with respect to other simpler techniques at acceptable costs.