UML associations: Reducing the gap in test coverage between model and code

Abstract

This paper addresses the overall problem of estimating the quality of a test suite when testing is performed at a platform-independent level, using executable UML models. The problem is that the test suite is often required to fulfill structural code coverage criteria. In the avionics domain it is usually required that the tests achieve 100% coverage according to logic-based coverage criteria. Such criteria are less effective when applied to executable UML models than when they are applied to code because the action code found in such models contains conditions in navigation and loops that are not explicit and therefore not captured by logic-based coverage criteria. We present two new coverage criteria for executable UML models, and we use an industrial application from the avionics domain to show that these two criteria should be combined with a logic-based criterion when testing the executable UML model. As long as the coverage is less than 100% at the model level, there is no point in running the tests at the code level since all functionality of the model is not yet tested, and this is necessary to achieve 100% coverage at the code level.