Obtaining Models for Test Generation from Natural-language-like Functional Specifications
Michael Esser, Peter Struß · 2007
The paper presents first results of a project that aims at a model-based tool for functional testing of control software for passenger vehicles. The objective is that this tool can be used in today’s engineering practice and, hence, the approach must not require costly changes in the current test generation process and not assume data and skills that do not exist in reality. Firstly, the input to test generation, i.e. the specification of functional requirements, exists mainly as natural language text, rather than in a formal language. Secondly, its output, i.e. a set of proposed tests, has to be justified and explained in terms that are comprehensible to the test engineers in order to allow them to inspect, revise, and complement it. We focus on design decisions that are motivated by this objective. The proposed solution offers a natural-language-template-based interface for acquiring software requirements. The content of the filled-in templates can be represented in propositional logic and temporal relations and form the model of the intended correct behavior. Models of potential faulty behaviors are generated from this OK model by a number of (types of) transformations. The fault types are defined mainly to match the intuition behind manually generated test cases and, hence, can deliver similar, but more systematic, test suites. This forms the basis for the intuitive justification of the tests and its manual postprocessing.