Searching for invariants using genetic programming and mutation testing

Invariants are concise and useful descriptions of a program's behaviour. As most programs are not annotated with invariants, previous research has attempted to automatically generate them from source code. In this paper, we propose a new approach to invariant generation using search. We reuse the trace generation front-end of existing tool Daikon and integrate it with genetic programming and a mutation testing tool. We demonstrate that our system can find the same invariants through search that Daikon produces via template instantiation, and we also find useful invariants that Daikon does not. We then present a method of ranking invariants such that we can identify those that are most interesting, through a novel application of program mutation.