Defect-based Quality Assurance with Defect Models

When performing quality assurance, software engineers are confronted with commonand recurring defects. To detect these defects, they typically exercise their knowl-edge and experience to create test cases or use test selection strategies encapsulatingknowledge in experience. For such test selection, there is at least anecdotal evidenceof its effectiveness. The problem is its usage of tacit knowledge leading to engineer-dependent test cases or test cases lacking a rationale as to why exactly they are effective.This thesis proposes a systematic and (semi-)automatic approach using defect modelsfor defect-based quality assurance as its major contribution. By capturing defect knowl-edge and experience of software engineers or inherent to test selection strategies indefect models, they are made explicit and described formally. Operationalizing defectmodels yields (semi-)automatic test case / check list generators directly targeting thedescribed defects for their effective and efficient detection.The defect models are accompanied by a defect model lifecycle framework to struc-ture the integration of defect-based quality assurance into existing quality assuranceprocesses. The lifecycle framework first contains activities to capture the tacit defectknowledge and experience by eliciting and classifying to arrive at an explicit libraryof common and recurring defects. The activities of formal description of defects indefect models and the operationalization of the defect models represent the methodapplication activities actually leading to the (semi-)automatic detection of the describeddefects. Finally, the activities of assessment and maintenance enable the evaluation ofeffectiveness and efficiency in a continuous support process.To comprehensively assess defect-based quality assurance based on defect models,we provide instantiations of the elicitation and classification as well as descriptionand operationalization activities in the lifecycle framework. For the elicitation andclassification activities, we present and evaluate the context-independent instantiationcalled DELICLA. Based on the results of DELICLA, we describe and operationalizedefect models on all levels of testing (i.e. 8Cage, OUTFIT and Controller Tester)and demonstrate their effectiveness and efficiency in detecting the described defects.These descriptions and operationalizations yield the second major contribution. Inaddition to the activity instantiations, we give rise to generic assessment criteria foroperationalizations and provide a framework for the maintenance of defect models.