Repair with On-The-Fly Program Analysis

This paper presents a novel automatic repair approach for incorrect programs. It applies formal methods and analyzes program behavior only on demand. We argue that this is beneficial, especially if exhaustive program analysis is infeasible. Our approach computes repair candidates and refines them based on counterexamples. It can be used with various verification techniques and specification formats to check a candidate's correctness. This includes test suites, model checkers verifying assertions, or even the user checking candidates manually, in which case no explicit specification is needed at all. We use concolic execution to analyze programs and SMT-solving to compute repair candidates. We implemented our approach in the open-source debugging environment FoREnSiC and present first experimental results.