AGEDIS Case Studies: Model-Based Testing in Industry
Ian Craggs, Manolis S. Sardis, Thierry Heuillard · 2003
During the three-year Automated Generation and Execution of Test Suites in Distributed Component-based Software (AGEDIS) project, five case studies were undertaken. Each one applied model-based testing methods and tools to a real industrial testing problem. The first two, at France Telecom and Intrasoft, took place before the AGEDIS tools were built to determine the successes or otherwise of existing technology. The final three, at IBM and again Intrasoft and France Telecom used the AGEDIS tools and method for comparison. The AGEDIS modelling language (AML) is based on UML, the de facto standard among modelling languages today. It was found that the act of modelling itself is beneficial, increasing the modeller’s understanding of the system under test, exposing inconsistencies that could otherwise have remained hidden. AML, being partly graphical, is good for communicating the structure of the model. Systems with interacting components tend to have complicated and in many cases non-deterministic behaviour. Test generation algorithms must therefore be able to deal with large models. Those based on full state space coverage suffer from combinatorial explosion; algorithms based on coverage of inputs and outputs fare better. The AGEDIS tools allow non-determinism to be dealt with adequately, but more needs to be done.