Towards Test-Driven Semantics Specification

Behavioral models are getting more and more important within the software development cycle. To get the most use out of them, their behavior should be defined formally. As a result, many approaches exist which aim at specifying formal semantics for behavioral languages (e.g., Dynamic Meta Modeling (DMM), Semantic Anchoring). Most of these approaches give rise to a formal semantics which can e.g. be used to check the quality of a particular language instance, for instance using model checking techniques. However, if the semantics specification itself contains errors, it is more or less useless, since one cannot rely on the analysis results. Therefore, the language engineer must make sure that the semantics he develops is of the highest quality possible. To help the language engineer to achieve that goal, we propose a test-driven semantics specification process: the semantics of the language under consideration is first informally demonstrated using example models, which will then be used as test cases during the actual semantics specification process. In this paper, we present this approach using the already mentioned specification language DMM.