Which Inputs Trigger My Patch?

Automated program repair (APR) has made significant progress in autonomously locating and fixing software faults. However, developers face uncertainty in trusting and validating patches, as the boundaries of the patch and the concrete inputs the patch fixes often remain unclear. In this paper, we introduce Avicennapatch, a novel tool de-signed to provide precise explanations of the input structures that are covered by the patch. Building on the Avicenna framework, Avicennapatch uses a differential testing strategy, executing both the original and patched program versions to identify inputs where behavior diverges. By leveraging ISLA constraints, Avicennapatch generates human-readable, grammar-based explanations that characterize patch-triggering inputs in terms of syntactic and semantic features. To refine these explanations, Avicennapatch iteratively generates new inputs and adjusts its hypotheses, ensuring that the explanations are both precise and generalizable. Our approach not only enhances transparency in APR by clarifying the operational boundaries of patches but also supports developers in evaluating the reliability and robustness of automated fixes across diverse input scenarios. Our evaluation on 12 patches for 7 different subjects shows that Avicennapatch can provide effective and general explanations.