Lo Gullstrand Heander's Licentiate Seminar

• Thesis title: Modelling and Improving Human-Centered Code Review
• Author: Lo Gullstrand Heander, Department of Computer Science, Lund University
• Faculty opponent: Senior Lecturer Francisco Gomes de Oliveira Neto, University of Gothenburg
• Examiner: Senior Lecturer Elizabeth Bjarnason, LTH
• Supervisor: Senior Lecturer Emma Söderberg, LTH
• Co-supervisor: Prof. Görel Hedin, LTH & Senior Lecturer Christofer Rydenfält, LTH
• Date and time: Thursday 11 September at 13:15
• Location: E:C, E-huset, LTH, Klas Anshelms väg 10/Ole Römers väg 3, Lund
• Link to download the thesis can be found here at the LU Research Portal

Abstract

Tool-based code review has been an established software engineering practice for at least a decade. However, while software development environments have improved significantly during this time with advanced features for code comprehension, refactoring, and AI support, code review tools have remained more static and are still centered around a two-way textual diff view with features similar to when the first code review tools were introduced. With the rapid development of artificial intelligence (AI), code review is at a crucial moment. It must adapt to meet the demands of a future where more and more AI generated code needs to be reviewed, while higher efficiency demands are placed on software engineering teams. More and more capable AI models will soon make it feasible to completely automate code review or offer sophisticated AI support to human code reviewers. Complete automation could potentially offer increased efficiency, but risk losing many of the interpersonal benefits. This gives researchers and software engineers reason to stop and reflect on what the purpose and benefits of code review are and how to best preserve these benefits in the future. In this thesis, I present a direction for modeling and improving human-centered code review, where code review tools are designed to support the human software engineer, adapt to their needs, and augment their capabilities. The contributions are a prototype for flexible code block comparisons developed using participatory design, an architecture for AI-supported code review, and a cognitive model of code review as decision-making (CRDM). Together, these contributions indicate one way toward the next generation of code review tools, practices, and processes: to use participatory design methodology, cognitive insights from the CRDM model, and AI agent-based architectures to improve code review while focusing on the needs of the human reviewers.