Crowdsolving Program Verification

In this effort, we consider the verification of properties in C (subset) programs. That is, we prove the validity of a pre/postcondition pair for a program, or demonstrate invalidity via an error trace. This is undecidable in general, and modern static analysis techniques struggle to reason about non-linear programs and programs with loops. To that end, we use abstraction for defining results of looping and non-linear computations. Furthermore, existing techniques do not take advantage of programmer’s insight, since the programmer receives little feedback from failed proof attempts. To that end, we ask user for facts that are relevant to the proof. Our technique is sound, modulo the correctness of verification tools we rely on. As a first step, existing static analysis and invariant generation techniques are deployed on the program. We try to prove the postcondition using these facts. If that fails, we compute an abstraction of the original program, and query user for facts about the abstract program. We check the validity of each user fact given the set of known proven facts. We again try to prove the postcondition using this new fact. This process iterates until we either discharge the proof, or demonstrate a bug. In summary, we hope to involve a crowd of semi-expert users in solving a verification problem. With game mechanics and proper incentive mechanisms in place, crowdsolving can help scale program verification.