The Btrfs file system may soon introduce formal patch submission guidelines in Linux Git, an effort aimed at handling a growing volume of contributions generated with the help of AI and large language models — along with what maintainers describe as "newbies without common senses are doing more harm than help."

Phoronix reported the proposal, which is still being shaped within the project and has not been settled. The complaint underneath it is not new: as the cost of producing a plausible-looking patch falls, the work of verifying one has not fallen with it.

Why review is the bottleneck

Btrfs is one of the more complex subsystems in the kernel tree. Reviewing a change means reasoning about on-disk formats, transaction semantics, failure modes and recovery paths — the kind of work where a patch can appear correct and still be wrong in ways that only surface under specific workloads or after a crash. That asymmetry is what makes AI-assisted submissions a particular burden: generating code is cheap, but the expensive step — checking whether it actually holds up — lands entirely on maintainers.

Maintainers voiced frustration that low-effort or AI-generated contributions are consuming review time without a corresponding gain in quality, with some observers characterising the influx as coming from people lacking the background to judge whether their own suggestions are sound. The quote circulated alongside the proposal is best read as an indication of review load rather than a policy position: it reflects discussion around the guidelines, not a rule the project has adopted.

What the guidelines aim to do

The mechanics are straightforward. A written document sets out what a good Btrfs submission should contain — a clear description of the problem, evidence that the change was tested, and enough context for a reviewer to understand the reasoning behind it. Patches that arrive with generic commit messages and no verification detail should, under the proposed guidelines, be easy to dismiss. The effect is to push a share of the verification cost back onto the submitter, and to give maintainers a concrete standard to point at when declining a contribution.

That matters beyond contributor etiquette. Btrfs underpins production deployments, and a merged change that quietly corrupts data or breaks recovery is a cost paid by downstream users, not by the person who generated it. The main outstanding issue is enforcement: the document still has to settle within the project, and its final wording may shape how strictly it is applied in practice.

A pattern beyond Btrfs

The proposal sits within a wider kernel-wide conversation about how projects absorb AI-assisted contributions, and the question is not confined to Btrfs. Whether written guidelines actually reduce noise, or simply become a box for submitters to tick, remains an open question — one the text itself acknowledges rather than resolves.

Teams anywhere that have adopted AI pair-programming face the same arithmetic: generation is cheap, review is not. The Btrfs proposal is a reminder that the control point is not the tool producing the code, but the process that decides what is worth a human's attention.

Source: Phoronix, https://www.phoronix.com/news/Btrfs-LLM-Newbies-Guidelines


Btrfs 檔案系統或將於 Linux Git 引入正式的 patch 提交指引,以應對藉助人工智能及大型語言模型產生、數量日增的投稿,同時亦針對維護者所形容的「缺乏常識的新手所造成的破壞多於幫助」的情況。

Phoronix 報道了這項建議;有關方案現仍在項目內醞釀,尚未定案。其背後的投訴並不新鮮:隨着製作一份看似合理的 patch 成本下降,核實一份 patch 所需的工夫卻並未同步下降。

為何審核成為瓶頸

Btrfs 是核心源碼樹中較為複雜的子系統之一。審核一項改動,意味着要推敲磁碟格式、交易語義、故障模式與復原路徑——在這類工作中,一份 patch 可以表面正確,卻在特定負載或系統崩潰之後才暴露出錯漏。正是這種不對稱,令 AI 輔助的投稿成為特別沉重的負擔:生成代碼成本低廉,但昂貴的一步——檢查它是否真的站得住腳——卻完全落在維護者身上。

維護者坦言,低投入或由 AI 生成的投稿正蠶食審核時間,卻沒有帶來相應的質素提升;部分觀察者形容這股投稿潮來自欠缺相關背景、無從判斷自身建議是否合理的人士。與建議一併流傳的那句話,與其視為政策立場,不如理解為審核負荷的指標:它反映的是圍繞指引的討論,而非項目已採納的規則。

指引旨在達成什麼

機制本身相當直接。一份書面文件列明一份優質 Btrfs 投稿應包含什麼——清楚說明問題、提供改動已測試的證據,以及足夠的背景資料,讓審核者理解背後的理據。按照建議中的指引,附帶千篇一律的 commit 訊息、又沒有驗證細節的 patch,應當易於退回。其效果是把部分核實成本推回投稿者身上,並讓維護者在拒絕投稿時有一個具體標準可依。

這超越了投稿禮儀的層面。Btrfs 是生產環境部署的基石,一項合併後悄悄損壞數據或令復原失效的改動,代價由下游用戶承擔,而非由生成該改動的人付出。目前主要的懸而未決問題在於執行:文件仍須在項目內定稿,其最終措辭或會左右日後實際執行的嚴格程度。

超越 Btrfs 的一種模式

該建議置身於核心社區更廣泛的對話之中,討論項目應如何吸納 AI 輔助的投稿,而問題並不限於 Btrfs。書面指引究竟能否真正減少噪音,抑或僅淪為投稿者用來打勾的方框,仍是未有定論的問題——文本本身承認這一點,但並未解決它。

世界各地採用 AI 結對編程的團隊都面對同一算式:生成廉價,審核不廉價。Btrfs 的建議提醒我們,控制點並非產出代碼的工具,而是決定什麼值得人力關注的流程。

來源: Phoronix, https://www.phoronix.com/news/Btrfs-LLM-Newbies-Guidelines

新聞來源 / Original News Source