Branching Processes for QuickCheck Generators (extended version) Agustín Mista Alejandro Russo John Hughes Universidad Nacional de Rosario Chalmers University of Technology Chalmers University of Technology Rosario, Argentina Gothenburg, Sweden Gothenburg, Sweden
[email protected] [email protected] [email protected] Abstract random test data generators or QuickCheck generators for short. In QuickCheck (or, more generally, random testing), it is chal- The generation of test cases is guided by the types involved in lenging to control random data generators’ distributions— the testing properties. It defines default generators for many specially when it comes to user-defined algebraic data types built-in types like booleans, integers, and lists. However, when (ADT). In this paper, we adapt results from an area of mathe- it comes to user-defined ADTs, developers are usually required matics known as branching processes, and show how they help to specify the generation process. The difficulty is, however, to analytically predict (at compile-time) the expected number that it might become intricate to define generators so that they of generated constructors, even in the presence of mutually result in a suitable distribution or enforce data invariants. recursive or composite ADTs. Using our probabilistic formu- The state-of-the-art tools to derive generators for user- las, we design heuristics capable of automatically adjusting defined ADTs can be classified based on the automation level probabilities in order to synthesize generators which distribu- as well as the sort of invariants enforced at the data genera- tions are aligned with users’ demands. We provide a Haskell tion phase. QuickCheck and SmallCheck [27] (a tool for writing implementation of our mechanism in a tool called generators that synthesize small test cases) use type-driven and perform case studies with real-world applications.