Automatic Testing through Planning

We describe a strategy to automatically test software built using preand postconditions. The strategy searches for valid routine calls: calls for which the preconditions are satisfied. If such calls fail (because the postcondition or another check is violated), we have found a bug. The testing strategy automatically builds a model of the software under test. The model is an abstract version of the semantics given by the preand postconditions. We use planning and learning to find a sequence of instructions that constructs the arguments to a valid routine call. The strategy is fully automatic and can be used to find bugs without intervention of the user. The generated test cases complement the unit tests that a designer may write by hand. We also illustrate the use of our experimental implementation on a non-trivial example.