Our industry collaboration with SAP Labs Korea, “Finding the Needle in the Crash Stack: Industrial-Scale Crash Root Cause Localization with AutoCrashFL,” has been accepted to ICSE-SEIP 2026 and will appear at the 48th ICSE in Rio 🇧🇷!

This work is an industry-oriented adaptation of our previous work, AutoFL [link][preprint], tailored specifically for large-scale crash debugging.

To address the practical challenges of (1) missing coverage information, (2) extremely large crash dumps, and (3) the need for efficient navigation of industrial C/C++ codebases, AutoCrashFL performs crash-dump parsing to reduce noise and narrow the search space, and grants the LLM direct access to both the source code and the clangd Language Server for precise code exploration.

Our results show that AutoCrashFL significantly outperforms a crash-stack-only baseline across all crash types, highlighting its strong industrial applicability.

Check out more details at our preprint! 👉 https://arxiv.org/abs/2510.22530