Abstraction and Automation for Reasoning about Complex Software

现代软件开发是一项复杂而混乱的业务。需求可能不完整。新的开发使用复杂的现有库、工具和组件，这些都可能会失败。多个开发团队按照不同的时间表进行，这使得很难保证他们的工件将按照预期“相互通信”。每次进行更改时，都需要执行昂贵的质量保证，以确保软件仍然能够按照预期工作。由于许多公司，特别是在汽车、电子、航空航天和国防领域，需要维护具有相似但不相同功能的软件产品变体家族（例如，支持不同类型的客户和/或软件平台），因此此类问题变得更加严重。我相信，工业上可扩展的自动推理和分析工具的可用性可以帮助解决上述许多问题，这些工具使用户能够回答有关其工件的基本问题。例如，他们可以确定所提议的更改的影响，询问关于他们的设计的“如果”问题，确定不同的功能是否会交互，检查组件的期望属性是否在任何环境中都能保持，等等。创建这样的工具需要在正式验证和软件工程方面的专业知识，因为许多挑战是在改进SE从业者的分析工具的可用性，在“正确”的抽象级别上执行分析，在开发生命周期中尚未获得完整信息的早期进行分析，在产品变体之间重用分析结果，等等。有效的和可扩展的自动化技术来解决各种复杂的软件工程问题。重点将放在对具有不完整信息的工件进行分析、处理产品线以及分析重用上。我打算利用我对验证工具的深入了解，如模型检查器和静态分析器，并努力建立正确的抽象来促进推理。该项目的成功将显著扩大自动化推理和分析在实践中的适用范围

Modern software development is a complex and messy business. Requirements are likely incomplete. New development uses complex existing libraries, tools and components that can fail. Multiple development teams proceed on different schedules, making it difficult to assure that their artifacts will "talk to each other" as intended. Every time a change is made, expensive quality assurance needs to be performed to ensure that software still does what it was supposed to. Such problems are made even more dire because many companies, especially in automotive, electronics, aerospace and defense domains, need to maintain families of software product variants with similar yet not identical functionality (e.g., supporting different types of customers and/or software platforms).***   I believe that many of the above problems can be aided by the availability of industrially scalable automated reasoning and analysis tools that enable users to answer essential questions about their artifacts. For example, they can determine the impact of a proposed change, ask "what if" questions about their designs, determine whether the different features are going to interact, check whether desired properties of a component would hold in any environment, etc. Creating such tools requires expertise both in formal verification and in software engineering, as many of the challenges are in improving usability of the analysis tools for SE practitioners, performing analysis at the "right" level of abstraction, enabling analysis earlier in the development lifecycle where complete information is not yet available, reusing results of analysis between product variants, etc.***   The goal of this project is to develop precise, effective and scalable automated techniques to solve a variety of complex software engineering problems. The emphasis will be on enabling analysis of artifacts with incomplete information, handling product lines, and analysis reuse. I intend to leverage my in-depth knowledge of verification tools such as model-checkers and static analyzers as well as strive to establish the right abstractions to facilitate reasoning. Success of this project will significantly expand the scope of applicability of automated reasoning and analysis in practice.**