Reproducible Automated Program Repair Is Hard -- Experiences With the Defects4J Dataset

• URL: http://arxiv.org/abs/2604.26674v1
• Date: Wed, 29 Apr 2026 13:46:13 GMT
• Title: Reproducible Automated Program Repair Is Hard -- Experiences With the Defects4J Dataset
• Abstract summary: We investigate the widely used benchmark Defects4J, which has according to Google Scholar over 1,800 citations.<n>We find that, when executing the test suites with strict requirements for APR settings, 180 (21.6 of the defects are not suitable for evaluation experiments.<n>An additional 59 (7.1 defects have test suites that are obviously under-specified, as deleting a single statement from the code base makes all test cases pass, although the human-written patch does not delete code.
• Score: 0.45835414225547183
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: In the research of automated program repair (APR), benchmark datasets consisting of known defects in combination with test suites that indicate the defects are of high importance. They allow for an evidence-based comparison of different APR approaches. In our own work on APR we found significant challenges when working with widely used defect datasets, which go beyond mere repeatability of defects via test cases. We summarize these identified challenges and related lessons learned to bring them to the attention of the APR community and quantify the potential impact of them. In particular, we investigate the widely used benchmark Defects4J, which has according to Google Scholar over 1,800 citations. It consists of 835 defects from 17 open-source Java projects; a hand-curated collection of defects, test suites that clearly indicate the defect, and human patches where any unrelated changes are removed. We find that, when executing the test suites with strict requirements for reproducibility in APR settings (beyond merely reproducing the defect via test cases), 180 (21.6 %) of the defects are not suitable for evaluation experiments. Further, we find that an additional 59 (7.1 %) defects have test suites that are obviously under-specified, as deleting a single statement from the code base makes all test cases pass, although the human-written patch does not only delete code. Our contributions are: a systematic collection of requirements for defect datasets for APR beyond traditional reproducibility of defects, a description of practical experiences and quantitative analysis of problems with the Defects4J dataset, as well as an implementation of an evaluation framework for APR tools for Java programs. This evaluation framework does stricter checking for indications of inadequate test suites, to avoid otherwise unnoticed problems in the test suite, such as flaky tests.

Related papers

• SWE-Gate: Passing Functional Tests Is Not Enough for Software Engineering Agents [63.65682461568621]
  We introduce SWE-Gate, a repository-level benchmark for software engineering agents that explicitly evaluates review constraint compliance alongside functional correctness.<n>SWE-Gate derives review constraints from real pull request review comments and synthesizes repository-level repair instances around these constraints.<n>Each instance provides separate functional and constraint tests, together with non-compliant and gold patches, enabling explicit separation between issue resolution capability and review constraint compliance.
  arXiv  Detail & Related papers  (2026-09-03T17:53:34Z)
• Beyond Fail-to-Pass: Iterative Hardening of Co-Generated Bug Reproduction Tests and Fixes [55.114939648705395]
  Large language models (LLMs) have made automated program repair increasingly practical for real-world bugs.<n>Bug reproduction tests (BRTs) help close this gap by turning a bug report into an executable, bug-specific signal.<n>We propose CoHarden, a co-generation framework that uses the Lax signal as an in-loop convergence criterion.
  arXiv  Detail & Related papers  (2026-07-22T07:30:07Z)
• Are Benchmark Tests Strong Enough? Mutation-Guided Diagnosis and Augmentation of Regression Suites [49.16055123488827]
  Insufficiently strong test suites can admit plausible yet semantically incorrect patches, inflating reported success rates.<n>We introduce STING, a framework for targeted test augmentation that uses semantically altered program variants as diagnostic stressors.<n>STING produces 1,014 validated tests spanning 211 instances and increases patch-region line and branch coverage by 10.8% and 9.5%.
  arXiv  Detail & Related papers  (2026-04-02T01:13:40Z)
• HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam [63.84155758655084]
  Humanity's Last Exam (HLE) has become a widely used benchmark for evaluating frontier large language models.<n>We introduce HLE-Verified, a verified and revised version of HLE with a transparent verification protocol and fine-grained error taxonomy.<n>We evaluate seven state-of-the-art language models on HLE and HLE-Verified, observing an average absolute accuracy gain of 7--10 percentage points.
  arXiv  Detail & Related papers  (2026-02-15T02:50:15Z)
• PaperAudit-Bench: Benchmarking Error Detection in Research Papers for Critical Automated Peer Review [54.141490756509306]
  We introduce PaperAudit-Bench, which consists of two components: PaperAudit-Dataset, an error dataset, and PaperAudit-Review, an automated review framework.<n>Experiments on PaperAudit-Bench reveal large variability in error detectability across models and detection depths.<n>We show that the dataset supports training lightweight LLM detectors via SFT and RL, enabling effective error detection at reduced computational cost.
  arXiv  Detail & Related papers  (2026-01-07T04:26:12Z)
• Specification-Guided Repair of Arithmetic Errors in Dafny Programs using LLMs [79.74676890436174]
  We present an APR tool for Dafny that uses formal specifications as oracles for fault localization and repair.<n>We localize faults through a series of steps, which include using Hoare logic to determine the state of each statement within the program.<n>Our tool achieves 89.6% fault localization coverage and GPT-4o mini yields the highest repair success rate of 74.18%.
  arXiv  Detail & Related papers  (2025-07-04T15:36:12Z)
• What You See Is What You Get: Attention-based Self-guided Automatic Unit Test Generation [3.8244417073114003]
  We propose Attention-based Self-guided Automatic Unit Test GenERation (AUGER) approach.<n>AUGER contains two stages: defect detection and error triggering.<n>It makes great improvements by 4.7% to 35.3% in terms of F1-score and Precision in defect detection.<n>It can trigger 23 to 84 more errors than state-of-the-art (SOTA) approaches in unit test generation.
  arXiv  Detail & Related papers  (2024-12-01T14:28:48Z)
• ContrastRepair: Enhancing Conversation-Based Automated Program Repair via Contrastive Test Case Pairs [23.419180504723546]
  ContrastRepair is a novel APR approach that augments conversation-driven APR by providing contrastive test pairs. We evaluate ContrastRepair on multiple benchmark datasets, including Defects4j, QuixBugs, and HumanEval-Java.
  arXiv  Detail & Related papers  (2024-03-04T12:15:28Z)
• Automated Test Case Repair Using Language Models [0.5708902722746041]
  Unrepaired broken test cases can degrade test suite quality and disrupt the software development process.<n>We present TaRGET, a novel approach leveraging pre-trained code language models for automated test case repair.<n>TaRGET treats test repair as a language translation task, employing a two-step process to fine-tune a language model.
  arXiv  Detail & Related papers  (2024-01-12T18:56:57Z)
• Automatic Generation of Test Cases based on Bug Reports: a Feasibility Study with Large Language Models [4.318319522015101]
  Existing approaches produce test cases that either can be qualified as simple (e.g. unit tests) or that require precise specifications. Most testing procedures still rely on test cases written by humans to form test suites. We investigate the feasibility of performing this generation by leveraging large language models (LLMs) and using bug reports as inputs.
  arXiv  Detail & Related papers  (2023-10-10T05:30:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.