Feature-Oriented Software Engineering

一个软件系统通常是根据它的组成特性来考虑的，其中每个特性都是一个可选的或增量的功能单元（例如，撤销、单词计数和跟踪更改是文字处理软件中的特性）。软件开发人员使用特性作为将工作分解为任务的标准，这些任务可以增量地、并行地或由第三方供应商执行。面向特征的软件工程支持软件系统的增量开发和快速演进。缺点是，整合看似正交的特征可能导致特征交互，其中一个特征意外地影响另一个特征的行为（例如，冲突的行动，紧急行为，违反假设）。交互是功能丰富的软件产品中常见的问题。一个值得注意的例子是：在两起波音737 MAX飞机坠毁事件中，一个故障的输入传感器导致多个特征交互，其中臭名昭著的MCAS特征压倒了与飞机俯仰相关的其他特征（包括飞行员控制的特征），迫使飞机俯冲。为了安全起见，软件工程师需要管理集成在一起时功能的行为。

A software system is often thought of in terms of its constituent features, where each feature is an optional or incremental unit of functionality (e.g., undo, word count, and track changes are features in word processing software). Software developers use features as a criterion for decomposing their work into tasks that can be performed incrementally, in parallel, or by third-party vendors. Feature-oriented software engineering enables incremental development and rapid evolution of software systems. The downside is that integrating seemingly orthogonal features can result in feature interactions where one feature unexpectedly affects the behaviour of another (e.g., conflicting actions, emergent behaviours, violated assumptions). Interactions are common problems in feature-rich software products. As a notable example: in both of the Boeing 737 MAX aircraft crashes, a faulty input sensor led to multiple feature interactions in which the infamous MCAS feature overpowered other features related to the aircraft's pitch (including pilot-controlled features) and forced the aircraft into a nose dive. To be safe, software engineers need to manage how features behave when integrated together. This proposal investigates two avenues of research that address different aspects of the Feature Interaction Problem. The first focuses on establishing a precise system-wide analysis to detect feature interactions in large feature-rich software systems. Novel advancements will be analyses that (1) examine actual software rather than models or test results (2) are comprehensive in the types of interactions they detect, (3) scale to large software products, and (4) are adequately precise. The second research avenue pursues a more holistic approach to manage feature interactions, by means of a software architecture that the features plug into. The architecture coordinates and constrains how features can execute and communicate, thereby avoiding or resolving certain classes of interactions. An architectural solution to managing interactions has the potential of ensuring safe integration of features, thereby reducing the number of residual interactions that the engineer must detect and resolve. The result will be technologies and HQP trained in feature-oriented designs and analyses that realize the improvements in developer productivity and product quality that feature-oriented software engineering has always promised.