Using SPIN for automated debugging of infinite executions of Java programs

This paper presents an approach for the automated debugging of reactive and concurrent Java programs, combining model checking and runtime monitoring. Runtime monitoring is used to transform the Java execution traces into the input for the model checker, the purpose of which is twofold. First, it checks these execution traces against properties written in linear temporal logic (LTL), which represent desirable or undesirable behaviors. Second, it produces several execution traces for a single Java program by generating test inputs and exploring different schedulings in multithreaded programs. As state explosion is the main drawback to model checking, we propose two abstraction approaches to reduce the memory requirements when storing Java states. We also present the formal framework to clarify which kinds of LTL safety and liveness formulas can be correctly analysed with each abstraction for both finite and infinite program executions. A major advantage of our approach comes from the model checker, which stores the trace of each failed execution, allowing the programmer to replay these executions to locate the bugs. Our current implementation, the tool TJT, uses Spin as the model checker and the Java Debug Interface (JDI) for runtime monitoring. TJT is presented as an Eclipse plug-in and it has been successfully x pub applied to debug comple