Bisection-based method cuts effort to identify duplicate compiler bugs by a third

Random testing has become invaluable for catching compiler flaws, but it creates a new problem: testers often discover the same bug repeatedly through different test cases. Sorting through these duplicates demands significant human effort, making bug deduplication a critical practical challenge for compiler developers.

Why current methods fall short#

Most existing deduplication systems rely on program analysis to extract features associated with bugs and match them across test cases. While theoretically sound, these approaches carry heavy computational costs and struggle to generalize across different compiler codebases and testing scenarios.

How bisection offers a simpler path#

Researchers exploring a new paper on arXiv investigated whether bisection—a straightforward debugging technique that pinpoints the specific code change responsible for a failure—could serve as the foundation for deduplication. Their analysis found that tracking which commit triggered a bug provides a meaningful signal for grouping duplicates, though it needs refinement to avoid false matches.

Introducing BugLens#

Building on these insights, the team developed BugLens, a deduplication method anchored in bisection but augmented with the detection of bug-triggering optimizations to catch edge cases and reduce errors.

Testing BugLens on five real-world datasets revealed substantial gains: it saved an average of 33.56% human effort compared to the state-of-the-art system Tamer, and 10.68% compared to D3, while identifying the same total number of unique bugs.

Why simplicity matters#

Because bisection is conceptually straightforward and requires no deep program analysis, BugLens offers genuine practical appeal for compiler teams. Its simplicity translates to better portability across different compiler projects and debugging contexts, making it a promising tool for scaling compiler quality assurance in real-world settings.