跳到主要內容

QGIS評測:開源地圖軟體封殺AI程式碼貢獻 | QGIS Review: Open-Source GIS Bans AI Code Contributions

By Kit 小克 | AI Tool Observer | 2026-10-13

🇹🇼 QGIS評測:開源地圖軟體封殺AI程式碼貢獻

QGIS,全球最大的開源地理資訊系統(GIS)軟體,10月宣布一項強硬決定:全面禁止AI協助的程式碼貢獻。這不是哪個企業的內部規定,而是一個靠志工維護、全球數十萬人在用的開源專案,被AI生成的PR逼到不得不劃紅線。

從「人在迴圈」到全面禁止

QGIS原本的政策(QEP408)其實還算友善:允許用AI輔助寫程式,但要求貢獻者標註「AI協助」並確保有人實際審查過(human in the loop)。理論上是個合理的折衷方案。

但核心開發者Nyall Dawson提出的新提案QEP428直接點破現實:這套標註制度在實務上完全失靈。於是QGIS社群投票通過,從現在起不接受任何AI協助的程式碼或文件貢獻,違者會被回退程式碼、甚至取消貢獻資格。

維護者撐不住的真實數字

  • 標註為「AI協助」的PR約有50個
  • 其中約40個被直接關閉、未合併
  • PSC成員Julien Cabieces直言:「外部貢獻者使用AI,給本就吃不消的維護者帶來了無法承受的壓力」

換算下來,失敗率超過八成。這代表大量AI生成的PR根本沒有通過基本品質把關,維護者花時間審查卻多半是白工。

政策具體管什麼

  • 範圍涵蓋程式碼、文件、任何提交到倉庫的素材
  • 貢獻者要主動確認提交內容沒有用LLM協助產生
  • 違規會面臨程式碼回退與貢獻者權限制裁
  • 但留了彈性:未來可能依個案鬆綁,不是鐵板一塊

這件事對我們有什麼意義

QGIS不是第一個喊累的開源專案,curl、各種Linux子系統過去都抱怨過AI生成issue和PR的「垃圾品質」問題。這波AI程式碼貢獻爭議反映一個很現實的落差:AI工具讓「生成一份看起來完整的PR」變得超容易,但「生成真正能用、理解上下文、符合專案慣例的程式碼」還是跟以前一樣難。中間這段差距,現在全部轉嫁給開源維護者。

如果你平常用AI寫程式碼要投PR到別的開源專案,誠實的建議是:生成完一定要自己完整測試、讀懂每一行、確認符合專案風格,再送出去。別把「AI幫我寫好了」當成「已經審查過了」。這次QGIS的禁令,說白了就是在懲罰那些把這兩件事搞混的人。

好不好用,試了才知道。


🇺🇸 QGIS Review: Open-Source GIS Bans AI Code Contributions

QGIS, the world's largest open-source geographic information system (GIS) project, made a hard call in October: a full ban on AI-assisted code contributions. This isn't a corporate policy memo — it's a volunteer-run open-source project used by hundreds of thousands of people, pushed to draw a hard line by a flood of AI-generated pull requests.

From "Human in the Loop" to a Full Ban

QGIS's previous policy (QEP408) was actually fairly reasonable: AI-assisted code was allowed, as long as contributors tagged it and confirmed a human had reviewed it. On paper, a sensible compromise.

But core developer Nyall Dawson's new proposal, QEP428, called out the reality: the tagging system simply didn't work in practice. The QGIS community voted it in, and as of now the project rejects all AI-assisted code or documentation contributions outright — violations trigger code reversion and loss of contributor rights.

The Numbers Behind Maintainer Burnout

  • Roughly 50 PRs were tagged as AI-assisted
  • About 40 of them were closed without merging
  • PSC member Julien Cabieces put it bluntly: "AI use from external contributors puts a massive and unbearable pressure on the shoulders of our maintainers, who already had a tough job before the era of LLMs"

That's an 80%+ rejection rate. Most of the AI-generated PRs simply didn't clear basic quality review, meaning maintainers spent real time reviewing work that was mostly wasted effort.

What the Policy Actually Covers

  • Scope includes code, documentation, and any material submitted to the repository
  • Contributors must actively confirm their submission wasn't produced with LLM assistance
  • Violations mean code reversion and sanctioned contributor rights
  • But there's flexibility built in — the rule may loosen case-by-case over time, not a permanent blanket ban

What This Means for the Rest of Us

QGIS isn't the first open-source project to hit this wall — curl and several Linux subsystems have already complained about low-quality AI-generated issues and PRs. This wave of AI code contribution friction exposes a real gap: AI tools make it trivially easy to produce a PR that *looks* complete, but producing code that actually works, understands project context, and follows conventions is just as hard as it ever was. That gap is now landing squarely on open-source maintainers' desks.

If you use AI to write code for PRs on other people's open-source projects, the honest advice is: actually test what it generates, read every line, and make sure it matches the project's conventions before submitting. Don't mistake "AI wrote it" for "it's been reviewed." QGIS's ban is, bluntly, punishing the people who confused those two things.

好不好用,試了才知道。(You won't know if it's good until you try it.)

Sources / 資料來源

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code