Semi-Automated Inference of Feature Traceability During Software Development: Master's Thesis

Abstract

Despite extensive research on software product lines in the last decades, ad-hoc clone-and-own development is still the dominant way for introducing variability to software systems. Therefore, the same issues for which software product lines were developed in the first place are still imminent in clone-and-own development: Fixing bugs consistently throughout clones and avoiding duplicate implementation effort is extremely diffcult as similarities and differences between variants are unknown. In order to remedy this, we enhance clone-and-own development with techniques from product-line engineering for targeted variant synchronisation such that domain knowledge can be integrated stepwise and without obligation. Contrary to retroactive feature mapping recovery (e.g., mining) techniques, we infer feature-to-code mappings directly during software development when concrete domain knowledge is present. In this thesis, we focus on the first step towards targeted synchronisation between variants: the recording of feature mappings. By letting developers specify on which feature they are working on, we derive feature mappings directly during software development. We ensure syntactic validity of feature mappings and variant synchronisation by implementing disciplined annotations through abstract syntax trees. To bridge the mismatch between change classification in the implementation and abstract layer, we synthesise semantic edits on abstract syntax trees. We show that our derivation can be used to reproduce variability-related real-world code changes and compare it to the feature mapping derivation of the projectional variation control system VTS by Stanciulescu et al.