Thread-sensitive points-to analysis for multithreaded Java programs

Abstract

Every running thread has its own thread context that consists of values of the fields of the target thread object. To consider the thread context in understanding the behaviors of concurrently running threads, we propose a thread-sensitive interprocedural analysis for multithreaded Java applications. Our thread-sensitive analysis exploits thread-context information, instead of the conventional calling-context information, for computing dataflow facts holding at a statement. The thread-sensitive analysis is highly effective in distinguishing dataflow facts for different threads, producing more precise dataflow information than non-thread-sensitive analysis. The analysis is also generally much more efficient than conventional (calling) context-sensitive analysis. It uses the target thread objects at a thread start site to distinguish different thread contexts. We give a thread-sensitive points-to analysis as an instance of thread-sensitive analysis. We have implemented it and give some experimental results. We discuss several possible applications of the analysis.