Detection and Impact Analyses of Feature Interactions

处理大型软件系统复杂性的常见策略是将系统分解为更容易处理的较小子系统。基于特性的分解是一种功能分解，其中特性是可以独立考虑、开发和演化的功能单元。因为特性提供了不同的功能，它们代表了不同的开发任务，这些任务可以增量地添加，由独立的团队并行执行，或者外包给第三方供应商。功能可能是可选的，甚至是可征税的，从而允许通过产品内衬进行产品的大规模定制。基于特征的分解的缺点是意想不到的特征交互，其中一个特征的行为干扰另一个特征的行为，导致冲突，紧急功能行为，并可能人为传染到其他特征和系统。该项目研究了四种互补分析，这些分析能够有效地推理特征交互和特征演化：(1)轻量级静态分析，使工程师能够对所有特征进行系统范围的分析，并识别可能导致安全危害的潜在特征交互；(2)可扩展的动态分析和(3)组合代码级分析，可以检查一个功能实现是否可以替代另一个功能实现，而不会影响整体系统行为；(4)产品线分析，使工程师能够同时有效地分析和验证多个特征配置，从而有可能向公众提供更多的定制产品选择。所建议的工作的预期结果是计算机支持的方法和技术的集合，用于分析不同软件开发工件中表达的特征之间潜在的和已实现的交互。我们将评估这些工具在检测交互方面的有效性，以及它们对大型软件系统的可伸缩性，并将部分地评估来自我们的工业伙伴的软件工件。

A common strategy for dealing with the complexity of large software systems is to decompose the system into smaller sub-systems that can be tackled more easily. Feature-based decomposition is a type of functional decomposition, in which features are units of functionality that can be considered, developed, and evolved independently. Because features provide distinct functionalities, they represent distinct development tasks that can be added incrementally, performed by separate teams in parallel, or outsourced to third-party vendors. Features may be optional or even tariffable, thereby allowing for mass customization of products through product-lining. The downside of feature-based decomposition is unexpected feature interactions, in which the actions of one feature interfere with those of another, leading to conflicts, emergent functional behaviours, and possibly artificial contagion to other features and systems.This project investigates four complementary analyses that enable efficient reasoning about feature interactions and feature evolution: (1) Lightweight static analyses that enable engineers to perform a system-wide analysis of all features, and identify potential feature interactions that can lead to safety hazards; (2) scalable dynamic analyses and (3) compositional code-level analyses that allow checking whether one feature implementation can be substituted for another without affecting the overall system behaviour; and (4) product-line analyses that enable engineers to analyze and verify multiple feature configurations simultaneously and efficiently, making it possible offer a larger selection of customized products to the public. The expected outcomes of the proposed work are a collection of computer-supported methods and techniques for analyzing potential and realized interactions among features expressed in different software-development artifacts. The tools will be evaluated with respect to their effectiveness in detecting interactions and their scalability to large software systems, and will be evaluated in part on software artifacts from our industrial partner.