Prioritized Static Slicing for Effective Fault Localization in the Absence of Runtime Information

Static slicing identifies the parts of a program that might affect another point in that program. Unfortunately, static slicing often produces large and imprecise results because of its conservative nature. Dynamic slicing can be a practical alternative, but it requires runtime information that might not be available, or be hard to obtain, or have low quality. To deal with the imprecision of static slicing, we present PRIOSLICE, a novel static approach that exploits the insight that not all statements reported by static slicing are equally likely to cause a program failure. PRIOSLICE first defines and solves a probabilistic model of program dependencies. In this model, some data dependencies are more likely to occur than others and control dependencies are less likely to propagate errors than data dependencies. PRIOSLICE then traverses the program backwards, just like static slicing, but prioritizing dependencies by likelihood. Our study of fault localization, presented in this paper, indicates that PRIOSLICE can help localize faults much more effectively than existing static-slicing approaches.