GNU policy blocks LLM-assisted Emacs patch

An Emacs performance patch targeting macOS was rejected from the emacs-devel mailing list after the submitter disclosed that GLM 5.2 had helped identify and refine the change. The work followed instrumentation, benchmarking, and codebase review focused on rendering issues, memory thrashing, virtual memory bloat, and loss of cache locality tied to macOS memory behavior.

Regexp processing became a key focus because it is used widely across Emacs. Earlier LLM-assisted investigations were described as weak, with patches that had minimal impact or misunderstood the problem, but GLM 5.2 produced reports that were reviewed, tested, benchmarked, and narrowed into a submission. The submitted patch was 92 lines long and included explanatory comments.

The rejection prompted criticism that the GNU rule rewards nondisclosure rather than reducing LLM use. The submitter said AI-assisted code should receive more scrutiny, not less, and challenged concerns about openness and legality, especially for open-weight models. Internal GNU discussion of the policy was also criticized as lacking transparency toward users.