David Airlie, the maintainer of the graphics driver stack in the Linux kernel, announced the implementation of an auxiliary automated patch review system that utilizes the AI platform Claude Opus 4.6 for checking changes. The system has been implemented as an experiment to assess the feasibility of using AI to optimize the workflows of employees at Red Hat.
AI review encompasses changes submitted for inclusion in the DRM (Direct Rendering Manager) subsystem. The publication of results is organized using mailing lists used in Linux kernel development. To avoid cluttering the main mailing list where the development of the DRM subsystem is conducted, a separate mailing list named drm-ai-reviews has been created for AI reports.
It is anticipated that patch authors may use AI reviews as initial feedback. The analysis of such reviews is not mandated, and authors are free to take AI reviews into consideration at their discretion. However, if AI reviews identify regressions that the patch authors have overlooked, it is possible that in the future, the analysis of AI reviews will be required more assertively.
Initially, to organize the reviews in Claude, Dave attempted to use a set of prompts proposed by Chris Mason that clarify the features of subsystem operations and protocols, allowing the AI assistant to better understand the context during reviews. Chris's prompt set was intended to identify regressions and analyze patches when they were sequentially applied to the current kernel code tree and to conduct final regression testing on the latest patch.
Ultimately, Dave had to modify and expand Chris's prompts, as it was necessary for the AI to review a series of patches as a whole and then conduct a deep analysis of each patch individually. Besides the current kernel code tree, the review was extended to the drm-next branch. Dave also prepared tools for integration with the Claude AI service and for organizing the mailing list.
Source: opennet.ru
