Why Do Pull Requests Go Silent? Uncovering the Barriers to Contribution Completion in Open-Source Code Review

Pull requests (PRs) are central to pull-based software development, enabling distributed code review and collaborative contribution in open-source projects. However, many PRs become inactive before integration and are eventually abandoned or closed, resulting in wasted effort for both contributors and maintainers. Although prior studies have examined PR abandonment and review delays, less is known about the specific contribution types, discussion-level barriers, and post-stalling collaboration patterns associated with PR inactivity. This study investigates what types of PRs are most common among stalled contributions, why inactivity occurs from authors’ and reviewers’ perspectives, and how stalling is associated with subsequent contributor and reviewer engagement. We analyzed 14,234 stalled PRs and 164,562 review comments from 19 popular GitHub repositories that used stale-bot workflows. We developed an LLM-based voting classifier to categorize stalled PRs by contribution type and combined quantitative analysis with qualitative coding of general and inline review discussions. Our findings show that feature-enhancement and issue-fixing PRs constitute the largest share of stalled contributions, together accounting for over 77% of classified stalled PRs. General review comments indicate that stalling is mainly associated with communication and coordination breakdowns, including missing interaction, delayed feedback, and unclear follow-up. Inline review comments further show that inactivity is not always caused by disengagement, as many PRs are blocked by technical and dependency-related issues such as failing checks, configuration problems, compatibility concerns, and environment mismatches. We also find that only 39.56% of contributors returned to submit subsequent PRs, while reviewer re-engagement with the same contributors was approximately 21%. These results suggest that PR inactivity is a socio-technical coordination problem involving communication, technical readiness, review ownership, and automation practices. Based on these findings, we provide recommendations for type-aware triage, clearer review feedback, explicit ownership of next actions, CI-related blocker management, and cause-aware stale-bot interventions