Challenges in Automating Error-Fixing Commit Classification for Linux Kernel and Cyber-Physical Systems

Automated classification of error-fixing commits in Git repositories presents significant challenges, particularly in complex software domains such as the Linux kernel and cyber-physical systems. This study applies natural language processing to categorize commits from these domains, aiming to identify recurring error patterns and assess the feasibility of scalable, cross-project classification. While the Linux kernel's extensive commit history enables the detection of common issues, such as concurrency flaws in system code, sparse and ambiguous commit messages limit accuracy, with classifiers achieving only moderate precision when distinguishing bug fixes from feature updates. In repositories of cyber-physical systems, domain-specific terminology and smaller datasets further degrade performance, yielding fragmented clusters that require manual validation. Key barriers include inconsistent developer annotation practices, domain shifts between system software and embedded systems, and the inherent complexity of linking textual descriptions to code changes. Although automated methods show promise for surfacing high-impact error patterns, their practical utility remains constrained by noisy data and the need for domain adaptation. This work underscores the limitations of current approaches while outlining pathways for integrating commit semantics with architectural context.