ISBN 978-3-8439-1221-1

978-3-8439-1221-1, Reihe Informatik

Malte Lochau
Model-Based Conformance Testing of Software Product Lines

357 Seiten, Dissertation Technische Universität Braunschweig (2012), Softcover, A5

Zusammenfassung / Abstract

Besides the inherent complexity of modern software systems, their growing diversity makes their efficient, yet quality-aware development very challenging. For instance, every modern automobile including its safety critical, embedded software system constitutes a product variant individually tailored to customer's configuration decisions. The upcoming ISO 26262 standard requires every automotive software system to be sufficiently tested, which is, in general, infeasible due to the high number of possible configurations. Software product line (SPL) engineering constitutes a promising approach for efficiently developing families of similar software systems on the basis of a generic core platform and explicit specifications of commonality and variability between product variants in terms of their features. Thereupon, SPL engineering propagates a comprehensive reuse of common development artifacts among variants. However, adopting reuse principles also to quality assurance techniques such as model-based testing is an open problem due to the crosscutting nature of feature-oriented variability specifications with respect to traditional software design artifacts, e.g., potentially causing unintended feature interactions. In this doctoral thesis a foundation is developed for efficient, yet reliable quality assurance of software product lines using model-based conformance testing. For test modeling UML state machines are applied. The operational semantics is formalized and test case specifications based on decorated LTS trace semantics define parameterized testing preorder relations as a basis for model-based conformance testing. The test model is extended for modeling behavioral variability to yield a reusable SPL test model. Explicit annotations of model elements denote selection conditions over feature parameters organized in a domain feature model constituting the variant space of the SPL. For the sound reuse of test artifacts among variants, two techniques are proposed. As a static approach, conditioned model slicing performs syntax-based, variability-aware reductions on SPL test models to support, e.g., feature interaction detection. As a dynamic approach, behavioral commonality among product variants is obtained on the basis of modal refinement notions for feature-annotated LTS models. The framework is implemented in a sample tool chain and experimental evaluations are performed by means of a case study from the automotive domain.