Fault localization is the key method in debugging. Among multiple fault localization approaches, Spectrum-based Fault Localization (SBFL), which is highly effective, receives much attention from academia and industries. Existing SBFL needs to instrument on every code entities to get the runtime coverage information. However, the cost is too expensive and therefore not practical in large-scale real-world software. Besides, if the faults of the program account for a rather small proportion, a thorough instrumentation is extremely inefficient.

In order to solve this problem, Prof. Zhiqiang Zuo and Prof. Xintao Niu explore the feasibility of partial instrumentation rather than complete instrumentation. Specifically, they proposed an abstraction-refinement-based statistical approach, which automatically filters out abstract suspiciousness of the target information. Therefore, entities needed to be instrumented and analyzed are greatly reduced. In addition, rigorous proof is given to secure this method, which indicates no loss on accuracy. Finally, experiments on two different scenario,offlineandonlinestatistical debugging, are conducted to compare this approach and traditional approach. The result shows that their approach significantly improves the effectiveness while avoiding impairing its detecting ability.

Their paperToward More Efficient Statistical Debugging with Abstraction Refinementhas published onACM Transactions on Software Engineering and Methodology(TOSEM)this year.