Observation-Based Modeling for Model-Based Testing

One of the single most important reasons that modeling and modelbased testing are not yet common practice in industry is the perceived difficulty of making the models up to the level of detail and quality required for their automated processing. Models unleash their full potential only through sufficient formality, and after being processed by tools. However, making sufficiently formal models is regarded by industry as laborious, expensive, and overall daunting. This article presents a solution for circumventing the manual modeling process by bootstrapping and devising the model automatically from observations, and using that model for test case generation. It describes a combination of existing techniques and tools, some of which are readily used in industry, to provide an iterative process with which test engineers can create a formal model of a component, use it to generate test stimuli, compare the observed component behaviour with the modeled behaviour, and refine the code of the model or the component quickly and easily, according to the issues detected. Moreover, this process turns modeling more into an activity akin to programming, favoured by developers. The technique is demonstrated on one case study, in which it finds issues on both the implementation and the specification