Improving Defect Management in Automotive Software Development, LiDeC - A Light-weight Defect Classification Scheme

Classification of software defects is a means to provide defect reports with a shared and well-defined structure. Quantitative analyses of the classification data facilitated by the shared structure are useful to industry practitioners and academic researchers. For practitioners, especially in large, complex or dynamic organizations, analyses can provide valuable information that characterize the development process, assist in identifying improvement opportunities, and provide one basis for predictions (e.g. product quality and resource needs). For researchers, classification data can facilitate evaluating effects of improved practices (e.g. new methods and tools) by analysing classified defect data before and after applying the hypothesized improved practice. Although recognized as a promising approach, there has to date been limited research reported on defect classification schemes—specifically on the efficiency of their application in industry, and on the reliability of the classification data. Efficient classification is desirable as it minimizes the time required to classify defects. Reliability of the classification data is important as it directly affects the reliability of conclusions drawn from analyses of the data. In this thesis, a defect classification scheme based on and compliant to the standard classification for software anomalies (IEEE Std. 1044) is described and evaluated. The classification scheme, LiDeC (Light-weight Defect Classification Scheme), was adapted to and applied in the development of automotive safety software. Through case studies and an experiment, LiDeC was evaluated with respect to its industrial applicability, efficiency and reliability. The results show that analyses of classification data can provide new and useful information about the effectiveness of current development practices. Applying a classification scheme adapted to the target organization results in analyses that are more directly relevant to that organization. Academic experimentation showed that classification schemes are easy to learn and to apply—even when lacking domain specific knowledge, the experiment subjects were able to arrive at rational classifications. The main contributions of the thesis include: the description of an adaptation of the standard classification scheme to a specific organization while maintaining standard compliance; an initial industrial evaluation of the applicability of the adapted classification scheme; a description of a methodology for comprehensively evaluating defect classification schemes; and finally, an investigation of current state-of-the-art with respect to defect classification, and a proposed roadmap for future research.