SysPro: Reproducing System-level Concurrency Bugs from Bug Reports

2026-01-01

Journal of Systems and Software

AI Summary

This work proposes a fully automated approach to reproduce system-level concurrency bugs, which are notoriously difficult to reproduce due to their non-deterministic nature and the scarcity of precise information in bug reports. Existing tools struggle to accurately control the interleaving order of system calls across threads. To address this, the method integrates natural language processing, information retrieval, and regular expression matching to extract critical system calls and their source code locations directly from unstructured bug reports for the first time. It then employs a classification-partitioning strategy to generate effective inputs and leverages dynamic source-code instrumentation to precisely orchestrate thread interleavings during execution. Experimental evaluation on real-world system benchmarks demonstrates that the approach efficiently and accurately reproduces multiple known concurrency bugs, substantially overcoming the limitations of traditional reproduction techniques in controlling execution order.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Search and Optimization: Sampling/Simulation-based Search
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization
• Security and Privacy: Large-scale security measurements

Problem

• system-level concurrency bugs
• bug reproduction
• bug reports
• non-deterministic bugs
• system call interleaving

Innovation

• concurrency bug reproduction
• system call interleaving
• bug report analysis
• dynamic instrumentation
• input generation