一句话：构建生命周期量子缺陷基准 QBugsSet，实证揭示量子特定缺陷修复更难且 LLM 自动修复率更低。
针对量子软件缺陷研究多局限于单一框架或单一质量属性、缺乏生命周期视角的问题，本文构建 QBugsSet 基准：从 Qiskit、Cirq、PennyLane 生态的真实 GitHub 修复提交中挖掘 2,429 个缺陷和 1,793 个真实负样本，系统比较量子特定缺陷与经典缺陷在生命周期分布、修复延迟、修复前空间传播、修复负担及 LLM 调试效果上的差异。结果表明，量子特定缺陷更集中于语义关键阶段，修复前持续更久（延迟中位数 17 天 vs. 7 天），更常需要多 hunk 协同修复（37.8% vs. 27.3%）；最佳前沿 LLM 的端到端完全修复率仅 3.82%，低于经典缺陷的 8.51%。
Despite growing evidence that bugs in quantum software exhibit distinctive characteristics, existing studies are typically scoped to individual frameworks or single quality attributes and do not examine how quantum-specific bugs behave across the broader software lifecycle. To address this gap, we construct QBugsSet, a lifecycle-based benchmark containing 2,429 bugs and 1,793 realistic negative samples, mined from real GitHub bug-fixing commits in the Qiskit, Cirq, and PennyLane ecosystems, and use it to conduct a comprehensive empirical study comparing quantum-specific and classical bugs across lifecycle distribution, bug-fix latency and pre-fix spatial propagation, repair burden, and LLM-based debugging effectiveness. We find that quantum-specific bugs are more concentrated in semantically critical stages, tend to persist longer before repair (median 17 vs. 7 days), and more often require coordinated multi-hunk repairs (37.8% vs. 27.3%). These lifecycle characteristics highlight a practical limitation of current automation: even the best frontier LLM achieves an end-to-end full repair rate of only 3.82% on quantum-specific bugs, compared with 8.51% on classical bugs. Overall, these findings provide evidence-based guidance for future debugging support that integrates stronger quantum-aware semantic reasoning with broader repository context.
@inproceedings{paperbot12426,
title = {How Quantum Bugs Live and Die: A Lifecycle-Based Empirical Study of Bugs in Quantum Software},
author = {Yasai Shi and Xiangxin Meng and Xiangjie Huang and Jian Zhang and Tianyu Wo and Xu Wang},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834439}
}