Reproduce “High-Quality Automated Program Repair” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Manish Motwani

Research from the University of Massachusetts Amherst proposes incorporating information from natural language software artifacts to improve the quality of patches generated by Automated Program Repair (APR) tools. It presents methods for objective patch quality evaluation, combines fault localization techniques to enhance accuracy, and demonstrates the automated generation of precise tests from natural language specifications, all aimed at addressing the prevalent patch overfitting problem.

Key insights

  • Patch overfitting is a pervasive issue in Automated Program Repair, with only 13.8% to 41.6% of plausible patches from state-of-the-art tools passing independent evaluation test suites.
  • Combining diverse fault localization techniques, such as Spectrum-Based (SBFL) and Information Retrieval-Based (IRFL), through rank aggregation (SBIR) substantially improves defect localization accuracy, identifying the buggy statement as top-1 for 18.1% of defects.
  • Natural language software specifications can be leveraged to automatically generate highly precise (98.4%) executable tests with oracles, covering crucial behaviors like exceptions and boundary conditions that are often missed by developer-written tests.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.