CaSS: Context-Aware Slice Summarization for Debugging Regression Failures
Sahar Badihi · 2025
Troubleshooting regression failures is a common yet time-consuming task for developers. Program slicing techniques aim to streamline this process by identifying relevant program statements, reducing the code developers need to inspect. By surveying over 50 practitioners from eight countries, we identified two main limitations in existing slicing-based approaches for debugging regressions. First, to minimize the number of statements presented to developers, these approaches often exclude contextual information essential for fully understanding the failure. Second, to maintain information propagation within the slice, they include lengthy computations that are unnecessary for understanding the failure. We further use these observations to propose a new slicing approach for debugging regression failures, named CASS. Our evaluation shows that CaSS better aligns with developers' needs while producing more concise slices than existing techniques.