Observation-based approximate dependency modeling and its use for program slicing

While dependency analysis is foundational to much program analysis, many techniques have limited scalability and handle only monolingual systems. We present a novel technique that aims approximate from relatively small number of perturbed executions. Our technique, MOAD (Modeling Observation-based Approximate Dependency), reformulates as the likelihood one element dependent on another (instead Boolean relationship). generates variants by deleting parts source code executing them while observing impact. thus infers model captures relationship between modification observation points. evaluate using slices obtained resulting probabilistic models. Compared existing observation-based backward slicing ORBS, requires 18.6% observations, are 12% larger average. Furthermore, we introduce notion forward slices. Unlike which inherently computes slices, MOAD’s model’s dependences can be traversed in either direction allowing us easily compute In comparison static slice, misses 0–6 lines (median 0), excessively 0–37 8) slice.