Alessandra Gorla, Assistant Research Professor, IMDEA Software Institute
Dynamic invariant detection allows mining of specifications from existing systems, but the quality of the resulting invariants depends on the executions observed: Unobserved behavior is not captured by dynamically inferred invariants, which may thus be unsound. Although this can be countered by producing additional executions with automated test generation techniques, it is crucial to generate new inputs that exercise relevant unobserved executions.
In this talk I will present an ongoing work aiming to produce a test suite that, despite the limited set of executions, can automatically discover many sound invariants for Java classes.