COPE+: A method for design and evaluation of product variants

Context: Successful software systems attract a large number of customers from diverse domains. These systems continuously evolve to accommodate feature requests of such diverse customer-base. At some point during system evolution, accommodating the features requests of all the customers in a single product is extremely challenging because of the diversity of these features. Additionally, this approach increases the complexity of the system architecture, hence degrading its quality properties such as modifiability, scalability, reliability, etc. Aim: It may be better to evolve the existing system into multiple product variants, each addressing a specific customers’ segment. The single system facing feature requests already has an existing architecture. To implement each one of the proposed new features, a set of architectural components have to be modified. The new features should be added in the existing system in such a way that the number of components being impacted is minimized. Method: Proposed method COPE+ uses density-based clustering of customers to create their segments and corresponding product variants. An evolutionary algorithm is used to generate a sequence for implementing new features such that the impact of adding these features in the existing system architecture is minimized. A promising product portfolio (containing multiple product variants) is determined by comparing all of the customers’ proposed product portfolios with results of feature impact analysis on the architecture. Product variants within the selected portfolio are ranked based on the similarity of their respective behaviours represented as statecharts with that of the existing system. Results: COPE+ is evaluated on jEdit, a popular open-source text editor. It recommended two product variants based on input from customers and architecture impact analysis. Results were validated using a popular domain analysis method. Conclusions: The proposed method searches a potentially very large solutions space to propose product variants that meet customers’ expectations while minimizing the changes in the existing system architecture. 1 We use the term product variant to refer to an individual product in a software product line or a product portfolio. 2 Granularity of an architectural component is not preset and can be defined at any convenient abstraction level such as packages, classes, etc.