Thesis Defense - Turner

Software users and developers have different perspectives. Users are focused on the problem domain, where a system's features are the primary concern. Developers are focused on the solution domain, where a system's life-cycle artifacts are key. This difference leads to difficulties producing successful software systems. At present, there is little understanding of how to narrow the gulf between the two perspectives.

This dissertation argues for establishing an organizing viewpoint, which we term feature engineering, to incorporate features into software engineering as first class objects. Features, acting as a bridge between the problem and solution domains, can begin to narrow the gulf.

This dissertation explores three aspects of features in software engineering. First, it establishes a framework for understanding software features. In this framework, we define the concept of feature as separate from feature implementation, its realization in the developed system. The framework also includes a model of features in software engineering. This model identifies relationships between a system's features and the major development artifacts. The final part of the framework is an evaluation of the potential use of feature information in the various software development activities.

The second and third aspects considered are the roles for features and feature relationships in configuration management and testing. In this dissertation, we argue that configuration management is pivotal for taking advantage of the various relationships that features structure. The ability of commercial configuration management systems to use information about features is evaluated, and our experiences using a prototype feature-based configuration management system are reported. Feature testing is identified as a useful extension to traditional testing. Feature tests are used to confirm that a feature implementation faithfully produces the desired behavior. Feature tests are also used to uncover feature relationships and interactions in the solution domain.