Section: New Results

Formal Models for Variability

Participants : Axel Legay, Rudolf Fahrenberg, Jin Hyun Kim.

This part of the report is more concerned with task 2. It studies variability aspects in the broad scope. As in the first year, we have decided to use the concept of product lines as a general framework to reason on the problematic.

The behaviour of a software system is often described in terms of its features, where each feature is a unit of functionality that adds value to the system. Feature-oriented software development (FOSD) is a software-development strategy that is based on feature decomposition and modularity. Features can be separate modules that are developed in isolation, allowing for parallel, incremental, or multi-vendor development of features. Feature orientation is particularly important in software product lines, where a family of related products is managed and evolved in terms of its features: a product line comprises a collection of mandatory and optional features, and individual products are derived by selecting among and integrating features from this feature set. A product line can be expressed as a single model, in which feature-specific behaviour is conditional on the presence of the feature in a product.

The downside of FOSD is that, although features are conceptualized, developed, managed, and evolved as separate concerns, they are not truly separate. They can interfere with each other, for example by trying to control the same variables, by issuing events that trigger other features, or by imposing conditions that suppress other features. Most of the early work on feature interactions focused on interactions that manifest themselves as logical inconsistencies, such as conflicting actions, nondeterminism, deadlock, invariant violation, or unsatisfiability. More recently, a more general definition of feature interaction has been introduced, in terms of a feature that is developed and verified to be correct in isolation but is found to behave differently when combined with other features, and it was shown how such behaviour interactions could be detected as a violation of bisimulation.

Another problem is that FTS models are monolithic models of full product lines. There is no means of modelling individual features and composing them into products or product-line models, or of specifying feature increments to an existing product-line model. As such, FTSs cannot be the mathematical basis for modelling technologies that support feature decomposition, composition, or incremental evolution of a product line.

Papers:

[11] (C)

Featured Transition Systems (FTSs) is a popular representation for software product lines: an entire product line is compactly represented as a single transition-machine model, in which feature-specific behaviour is guarded by feature expressions that are satisfied (or not) by the presence or absence of individual features. In previous work, FTS models were monolithic in the sense that the modeller had to construct the full FTS model of the product line in its entirety. To allow for modularity of FTS models, we propose here a language for extending an existing FTS model with new features. We demonstrate the language using a running example and present results about the language's expressivity, commutativity of feature extensions, feature interactions, and resolution of such interactions.

[12] (C)

We suggest a method for measuring the degree to which features interact in feature-oriented software development. To this end, we extend the notion of simulation between transition systems to a similarity measure and lift it to compute a behaviour interaction score in featured transition systems. We then develop an algorithm which can compute the degree of feature interactions in a featured transition system in an efficient manner.