Flexible feature binding in software product lines

  • Authors:

  • Marko Rosenmüller;Norbert Siegmund;Sven Apel;Gunter Saake

  • Affiliations:

  • School of Computer Science, University of Magdeburg, Magdeburg, Germany;School of Computer Science, University of Magdeburg, Magdeburg, Germany;Department of Informatics and Mathematics, University of Passau, Passau, Germany;School of Computer Science, University of Magdeburg, Magdeburg, Germany

  • Venue:
  • Automated Software Engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

A software product line (SPL) is a family of programs that share assets from a common code base. The programs of an SPL can be distinguished in terms of features, which represent units of program functionality that satisfy stakeholders' requirements. The features of an SPL can be bound either statically at program compile time or dynamically at run time. Both binding times are used in SPL development and have different advantages. For example, dynamic binding provides high flexibility whereas static binding supports fine-grained customizability without any impact on performance (e.g., for use on embedded systems). However, contemporary techniques for implementing SPLs force a programmer to choose the binding time already when designing an SPL and to mix different implementation techniques when multiple binding times are needed. We present an approach that integrates static and dynamic feature binding seamlessly. It allows a programmer to implement an SPL once and to decide per feature at deployment time whether it should be bound statically or dynamically. Dynamic binding usually introduces an overhead regarding resource consumption and performance. We reduce this overhead by statically merging features that are used together into dynamic binding units. A program can be configured at run time by composing binding units on demand. We use feature models to ensure that only valid feature combinations can be selected at compile and at run time. We provide a compiler and evaluate our approach on the basis of two non-trivial SPLs.