CAREER: Transforming Peer Code Review Environments for Code Learning and High-Quality Feedback

同行代码评审（CR）是大多数开源和商业软件开发组织的强制性软件验证实践。在这种实践中，一个或多个对等体在将代码更改集成到项目的存储库之前检查并批准代码更改。由于开发人员每天都在CR任务上花费大量精力，因此提高CR效率是这些组织的优先事项。限制CR有效性的挑战包括：i）在有限的时间和代码上下文下的代码学习困难，ii）参与者对混淆建议的误解，以及iii）由于不尊重的反馈而引起的人际冲突。在短期内，这些挑战增加了所需的努力，延迟了结果，增加了拒绝的可能性，并使参与者感到沮丧。从长远来看，这些挑战会降低软件质量，导致参与者之间的冲突，使不适当的书面审查目标失去动力，对新人的入职构成障碍，对少数群体造成不成比例的影响，甚至导致长期开发人员永久离开。尽管有几项研究证实了这些短期和长期后果，但这些挑战的实际解决方案仍然不存在。 该项目将努力使用经验方法、机器学习和自然语言技术来改进代码审查，以生成用于代码审查的工具。新的知识和工具将被用作教育平台，为学生和新程序员提供支持。该项目将通过在课堂和课程开发中使用工具，将研究融入教育中，使用课堂环境来了解如何在专业代码审查生态系统中支持开发人员。该项目的总体目标是转换代码审查工具和工作流程，以解决参与者在理解审查中的代码方面的挑战，并以明确和建设性的语言与他人交流这种理解。该项目将根据经验为上述三个CR挑战中的每一个制定理论框架，以描述根本原因和潜在的缓解解决方案。该实验平台将把CR工具转变为一个在线学习环境，其中集成了额外的认知支持工具，以支持未满足或部分满足的信息需求，并及时指导，以帮助CR参与者明确和建设性地进行沟通。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

Peer Code Review (CR) is a mandatory software verification practice among most Open Source and commercial software development organizations. In this practice, one or more peers inspect and approve a code change before integrating it into a project's repository. As developers spend significant effort daily on CR tasks, improving CR effectiveness is a high priority for these organizations. Challenges limiting CR effectiveness include i) code learning difficulties with limited time and code context, ii) misunderstandings among the participants over confusing suggestions, and iii) interpersonal conflicts due to disrespectful feedback. In the short term, these challenges increase required efforts, delay the outcomes, increase the likelihood of rejections, and frustrate the participants. In the long term, these challenges degrade software quality, cause conflicts among the participants, demotivate an inappropriately written review's target, pose barriers to newcomers' onboarding, disproportionately impact minorities, and even cause long-term developers to leave permanently. Despite several studies confirming these short and long-term consequences, practical solutions to these challenges remain nonexistent. The project will work to improve code reviews using empirical methods, machine learning and natural language techniques to produce tools to be used in code reviews. The new knowledge and tools will be used as an educational platform that will support students and new programmers. The project will integrate the research into education by using the tools in classes and curriculum development, using the classroom setting to gain understanding of how to support developers in professional code-review ecosystems. The overarching goal of this project is to transform code review tools and workflows to address participants' challenges in understanding the code under review and communicate that understanding with others in unambiguous and constructive languages. This project will empirically develop a theoretical framework for each of the three aforementioned CR challenges to characterize the root causes and potential mitigating solutions. The experimental platform will transform CR tools into an online learning environments that integrate additional cognitive support tools to support unmet or partially met information needs and just-in-time coaching to assist CR participants in communicating unambiguously and constructively.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.