Usually, companies meet different customer needs in a particular domain by developing variants of a software product.This is often performed by ad-hoc copying and modifying of various existing variants to fit purposes of new one.As the number of product variants grows, such an adhoc development causes severe problems to maintain these variants.Software Product Line Engineering (SPLE) can be helpful here by supporting a large-scale reuse systematically.SPL architecture (SPLA) is a key asset as it is used to derive architecture for each product in SPL.Unfortunately, developing SPLA from scratch is a costly task.In this paper, we propose an approach to contribute for recovering SPLA from existing product variants.This contribution is two-fold.Firstly, identifying common features and variation points of features of a given collection of product variants.Secondly, exploiting commonality and variability in terms of features to identify mandatory components and variation points of components as an important step in this recovering process.To evaluate the proposed approach, we applied it to two case studies.The experimental results bring evidence the effectiveness of our approach.

Software Product Line Engineering (SPLE) is a systematic reuse approach to develop a short time-to-market and quality products, called Software Product Line (SPL). Usually, a SPL is not developed from scratch but it is developed by reusing features (resp. their implementing source code elements) of existing similar systems previously developed by ad-hoc reuse techniques. The features implementations that are reused may be changed for developing new products (SPL) using SPLE. Any code element can be a part of (shared by) different features implementations; modifying one feature's implementation can thus impact others. Therefore, feature-level Change Impact Analysis (CIA) is important to predict affected features for change management purpose. In this paper, we propose a feature-level CIA approach using Formal Concept Analysis (FCA) applied to SPL evolution. In our experimental evaluation using three case studies of different domains and sizes, we show the effectiveness of our technique in terms of the most commonly used metrics on the subject.