Technical debt in pull requests : insights from static code analysis and human reviews.

Abstract

Technical Debt (TD) refers to quick technical solutions that expedite delivery but degrade software quality over time, making systems harder to maintain, evolve, and test. Previous studies have mainly examined TD’s long-term evolution through commits or releases, focusing on either automated detection or code reviews. Yet little is known about its short-term dynamics—how developers and reviewers manage TD during pull requests (PRs), the main entry point for code integration. To address this, we conducted an empirical study on 12 Apache Java projects, analyzing TD in PRs from two angles: (i) a quantitative study using SonarQube to assess TD in 2,056 merged PRs, examining debt trends (reduced, unchanged, or increased), time influences, and frequently overlooked or resolved issue categories; and (ii) a qualitative study of 250 PRs and 929 review threads to investigate how reviewers and developers address TD, identifying triggers for TD discussions and the nature of suggestions for addressing TD. Our findings show TD is prevalent, appearing in 96.26% of PRs (tool-detected) and 84.4% (code reviews), with 76% of review threads discussing TD. TD evolution follows a 1:2:1 ratio (reduced:unchanged:increased). Common issues include code duplication, naming inconsistencies, cognitive complexity, and poor logging, with some deemed tolerable contextually. While 96% of TD discussions include suggestions for addressing TD, over 80% are fixed, though nearly 40% lack explicit developer confirmation, rising to 90% for unresolved issues. These insights can guide practitioners in managing TD, support ASAT tool developers in refining or designing tools, and provide a foundation for future research using our methodology and datasets.