Inappropriate reviewer assignments can undermine the benefits of code review, especially when stale reviewers, i.e. those who are no longer active, are recommended. We study the prevalence of stale recommendations across five reviewer recommendations tools (LearnRec, RetentionRec, cHRev, Sofia, WLRRec) on three large open-source projects. On average, 12.59% of incorrect recommendations are stale. Top reviewers often dominate these, with the top-3 accounting for half of stale cases in 15.31% of instances. Some reviewers are suggested up to 7.7 years after leaving. We propose a strategy to filter based on recent activity, reducing staleness by 21.44%–92.39%, though it may shift load to active reviewers.

Speakers:
Farshad Kazemi
MLOps/SE researcher - Arteria AI