Custom-Tailored Product Line Extraction
Wille, David · Digitale Bibliothek Braunschweig (Verbundzentrale Göttingen (VZG)) · 2019
Industry faces an increasing number of challenges regarding the functionality, efficiency and reliability of software. A common approach to reduce the linked development effort and respective costs are model-based languages, such as Matlab/Simulink and statecharts. While these languages help companies during development of single systems, the high demand for customized software is an increasing challenge. As a result, variants with high similarity and only slight differences have to be developed in an efficient way. As reimplementation of complex functionality for each variant is no option, copies of existing solutions are often modified for new customers. In the short-run, this so-called clone-and-own approach allows to save costs as existing solutions can easily be reused. However, this approach also involves risks as the relations between the copied systems are rarely documented and errors have to be fixed for each variant in isolation. Thus, with a growing number of potentially large system copies, the resulting maintenance effort can become a problem. To overcome these problems, this thesis contributes an approach to semi-automatically migrate existing model variants to software product lines. These product lines allow to generate all variants from the identified reusable artifacts. As industry uses a variety of different modeling languages, the focus of the approach lies on an easy adaptation for different languages. Furthermore, the approach can be custom-tailored to include domain knowledge or language-specific details in the variability identification. The first step of the approach performs a high-level analysis of variants to identify outliers (e.g., variants that diverged too much from the rest) and clusters of strongly related variants. The second step executes variability mining to identify corresponding low-level variability relations (i.e. the common and varying parts) for these clusters. The third step uses these detailed variability relations for an automatic migration of the compared variants to a delta-oriented software product line. The approach is evaluated using publicly available case studies with industrial background as well as model variants provided by an industry partner.