Predicting Runtime Data Dependences for Slice Inspection Prioritization

Data dependences are building blocks for techniques such as program slicing and, thus, are useful for program comprehension and debugging. Unfortunately, static analysis identifies many data dependences that are infeasible or unlikely to occur at runtime. Dynamic analysis, in contrast, does not identify all data dependences that can occur in a program. To address this inaccuracy, we present a new static analysis that predicts the likelihood of occurrence of data dependences to help engineers prioritize their inspection of those dependences. Our preliminary results indicate that our analysis, while having room for improvement, has a promising predictive power.