CAREER: Autonomous Targeted Software Verification

当今社会高度依赖于基于软件的系统，并且极易受到软件缺陷的影响。查找和修复软件缺陷的工作消耗了软件成本的很大一部分。现在，在软件系统的开发和操作过程中不断地应用缺陷发现技术，审查每一个代码更改（现代代码审查，简称MCR），并持续地测试软件的缺陷（持续集成，简称CI）是很常见的。不幸的是，这些持续的缺陷发现工作为软件开发人员带来了高昂的成本，并且它们提供的成功有限。这个项目的目标是减少软件开发人员花费在代码审查和持续集成上的无效的手工工作，同时尽可能多地保留他们富有成效的任务。为了达到这个目标，这个项目将开发一些技术来自动地对评审和集成操作进行优先级排序，对那些更容易发现缺陷的操作给予更高的优先级。这个项目将促进对哪些软件变更是有风险的，哪些变更更容易被开发人员接受，以及什么使开发人员信任自动目标缺陷查找技术的理解。它还将产生许多技术和工具，使软件工程师能够在更短的时间内发现更多的软件缺陷。该项目将通过提高软件可靠性和降低成本来造福社会。该项目将对软件工程师进行访谈，以了解影响自动化目标MCR和CI技术采用的人为因素。该项目致力于使用机器学习和基于搜索的算法实现三个目标：通过确定哪些代码部分不太可能改进，减少执行MCR和CI的代码更改的大小；自动执行由MCR和CI产生的操作。该项目还将研究如何生成使用这些技术做出的决策的自动解释。长期的愿景是提供一个集成的系统，自动减少MCR和CI任务的数量和大小，自动执行其中的一些任务，并向软件工程师解释其自动决策。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Today's society is highly dependent on software-based systems and highly vulnerable to the consequence of software defects. A high percentage of software costs are consumed by efforts to find and fix software defects. It is now common to apply defect-finding techniques continuously throughout development and operation of software systems, reviewing every code change (Modern Code Review, abbreviated MCR) and testing the software for defects continuously (Continuous Integration, abbreviated CI). Unfortunately, these continuous defect-finding efforts incur high cost for software developers, and they provide limited success. The goal of this project is to reduce the unfruitful, manual effort that software developers spend on code reviews and continuous integration, while keeping as many of their fruitful tasks as possible. To achieve this goal, this project will develop techniques to automatically prioritize review and integration actions, giving higher priority to those that are more likely to find defects. This project will advance the understanding of what software changes are risky, which ones are better accepted by developers, and what makes developers trust automatically-targeted defect-finding techniques. It will also produce many techniques and tools to enable software engineers to find more software defects in less time. This project will benefit society by improving software reliability, as well as reducing its cost.The project will conduct interviews to survey software engineers to understand the human factors that would impact the adoption of automated targeted MCR and CI techniques. The project works toward the achievement of three objectives using machine learning and search-based algorithms: reduce the size of the code changes for which MCR and CI get executed, by determining which code sections are unlikely to improve; automatically carry out actions resulting from MCR and CI. The project will also investigate how to generate automated explanations of the decisions made using these techniques. The long-term vision is to provide an integrated system that automatically reduces the number and size of MCR and CI tasks, automatically performs some of them, and explains its automated decisions to software engineers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.