Contemporary Software Peer Review: Modern practices, fault prediction, and extraction of design decisions

同行评审被公认为软件工程的最佳实践已经超过35年。它基本上是一种质量保证技术，依赖于专家互相检查工作。过于正式的过程限制了评审实践的采用，并且不能提高评审的有效性。当代的同行评审包括一系列技术，这些技术允许评审过程适合开发团队，并且是轻量级的，并且可以由管理者测量。从业人员推动了支持当代同行评议的工具的发展。虽然这种风格的评审已经在软件公司中以各种形式被采用，但还没有系统的研究。此外，从同步的基于纸张的评审到异步的工具支持的评审的转变留下了丰富的关于软件工件的复杂讨论的档案。这种归档的非结构化性质意味着它还没有被用于故障预测和系统历史的总结。我开发了一个工具ACER，它可以从自由格式的文本和代码片段中准确地提取代码元素，比如类和方法。我的工作将（1）系统地向从业者和研究人员描述当代审查;（2）帮助开发人员隔离系统的脆弱区域并理解系统演化。本文的研究目标是：1.1：对AMD和Microsoft等软件公司的同行评审进行系统的调查，1.2：在新环境下调整和验证当代评审（DND），1.3：确定评审制度的变化对软件质量的影响，2.1：使用ACER挖掘评审讨论的内容，以预测评审工作应该集中在哪里，2.2：从评审讨论中提取系统演进和设计决策。我的首要目标是将软件开发作为一门工程学科来推进，并帮助政府和公司按时、按预算生产高质量的软件。

Peer review has been acknowledged as a software engineering best practice for over 35 years. It is fundamentally a quality assurance technique that relies on experts checking each other's work. Overly formal processes limit adoption of review practices and do not increase review effectiveness.Contemporary peer review encompasses a series of techniques that allow the review process to fit the development team and to be both lightweight and measurable by managers. Practitioners have driven the development of and tools that support contemporary peer review. While this style of review has been adopted in various forms in software firms, there has been no systematic study of it. Furthermore, the shift from synchronous paper-based review to asynchronous tool-supported review has left a rich archive of sophisticated discussion of software artifacts. The unstructured nature of this archive has meant that it has not been used for fault prediction and summarization of system history. I have developed a tool, ACER, that can accurately extract code elements, such as classes and methods, from freeform text and code snippets. My work will (1) systematically describe contemporary review to practitioners and researchers; (2) aid developers in isolating the vulnerable areas of a system and understanding system evolution. My research objectives are to1.1: conduct a systematic investigation of peer review in software firms, including AMD and Microsoft,1.2: adapt and validate contemporary review in new settings (DND),1.3: determine the impact of variations in review regime on software quality,2.1: use ACER to mine the content of review discussion to predict where review effort should be focussed,2.2: extract system evolution and design decisions from review discussions. My overarching goal is to advance software development as an engineering discipline and to help government and firms produce high-quality software on time and on budget.