Property-based testing for functional programs

Licentiatavhandling, 2011

This thesis advances the view that property-based testing is a powerful way of testing functional programs, that has advantages not shared by traditional unit testing. It does this by showing two new applications of property-based testing to functional programming as well as a study of the effectiveness of property-based testing. First, we present a tool, QuickSpec, which attempts to infer an equational specification from a functional program with the help of testing. The resulting specifications can be used to improve your understanding of the code or as properties in a test suite. The tool is applicable to quite a wide variety of situations. Second, we describe a system that helps to find race conditions in Erlang programs. It consists of two parts: a randomised scheduler to provoke unusual behaviour in the program under test and allow replayability of test cases, and a module that tests that all of the functions of an API behave atomically with respect to each other. Finally, we present an experiment we carried out to compare property-based testing against test-driven development. The results were inconclusive, but in the process we developed a black-box algorithm for automatically grading student programs by testing, by inferring for each program a set of bugs that the program contains.