Isolating and Understanding Program Errors Using Probabilistic Dispute Model

Automated software debugging can have a signifi-cant impact on the cost and quality of software development and maintenance. In recent years, researchers have invested a considerable amount of effort in developing automated techniques, and have demonstrated their effectiveness in helping developers in certain debugging tasks by pinpointing faulty statements. But there is still a gap between examining a faulty statement and understanding root causes of the cor-responding bug. As a step in this direction, we believe good developers have defensive programming in minds and software debugging is a process in search of arguments about why a statement is faulty. Therefore, a fault localization problem is rephrased as a dispute game between statements involved in successful runs and failing runs. A statement is OK if it can always provide arguments against other's blames, whereas a less defensive statement is thought to be faulty. In doing so, we propose a probabilistic dispute graph which is built upon dynamic dependencies between statements and statistics of program runs. Using such a graph, we put executed statements in dispute, compute acceptable statements, and thus figure out faulty statements if they have not strong arguments about their correctness. For empirical purpose, we carry out experiments on the well-known Siemens benchmark, and conclude that our approach not only casts new light on the causes of bugs in various cases, but also is statistically more effective in fault localization than competitors like Tarantula, SOBER, CT and PPDG.