Random Testing For Language Design

Property-based random testing can facilitate formal verification, exposing errors early on in the proving process and guiding users towards correct specifications and implementations. However, effective random testing often requires users to write custom generators for well-distributed random data satisfying complex logical predicates, a task which can be tedious and error prone. In this work, I aim to reduce the cost of property-based testing by making such generators easier to write, read and maintain. I present a domain-specific language, called Luck, in which generators are conveniently expressed by decorating predicates with lightweight annotations to control both the distribution of generated values and the amount of constraint solving that happens before each variable is instantiated. I also aim to increase the applicability of testing to formal verification by bringing advanced random testing techniques to the Coq proof assistant. I describe QuickChick, a QuickCheck clone for Coq, and improve it by incorporating ideas explored in the context of Luck to automatically derive provably correct generators for data constrained by inductive relations. Finally, I evaluate both QuickChick and Luck in a variety of complex case studies from programming languages literature, such as information-flow abstract machines and type systems for lambda calculi.