Masters Thesis Presentation - Monica Daniella Homescu

Title: Who Reviews Matters: A Case Study of Post-Release Defects at Mozilla

Abstract:

Code review is a standard quality assurance step in software development. Most research has focused on which reviewers should be chosen for a change and on what makes a review effective at catching defects. Less attention has been paid to the assignment choices authors make themselves. Within Mozilla, authors can assign a revision to an individual reviewer or to a group. Prior work has found that group-reviewed revisions are less likely to introduce post-release regressions, but has not examined what leads an author to make that choice or whether the result holds once other characteristics of the revisions and the people involved are considered. This study investigates both. The analysis covers revisions submitted to the Firefox repository between 2018 and 2026, using code review data from Phabricator and bug tracking data from Bugzilla. The results show that reviewer assignment is driven mainly by prior habits, with an author's historical use of group review being the strongest predictor of their current choice and component-level history playing the same role for authors submitting their first revision. Across all analyses, group review is linked to a lower regression risk, with the estimated average difference around 5 percentage points. Even after accounting for revision size, author experience, reviewer engagement and component history, the association with group review holds. The results point to a broader picture in which regression outcomes depend on a combination of factors and the habit-driven nature of assignment raises questions about how well revisions are being matched to the review strategy that suits them best.

Supervisors: Ayushi Rastogi, Paris Avgeriou, Marco Castelluccio