reviewer-bot: workflow_dispatch reads default-branch code for explore tools (set REVIEW_CONTENT_ROOT)
还没有人认领这个 Issue。
评估
- 难度
- 2/5
- 预计耗时
- 1-3 小时
- 新手友好度
- 84/100
调研方向
从 reviewer-bot.yml 开始,结合检查 review_bot/run_review.py 中 _content_root() 的行为,检查 workflow_dispatch 路径。验证如何获取 HEAD_SHA,以及在不更改受信任的主 checkout 的情况下,应如何连接单独的 checkout 和 REVIEW_CONTENT_ROOT。完成标准是手动 dispatch 能够探索 PR head,同时 pull_request 的行为保持不变。
由索引模型根据 Issue 内容生成。
描述
Follow-up from PR #862 review (thread on reviewer-bot.yml).
Problem
On the workflow_dispatch (manual) trigger, reviewer-bot.yml does a single actions/checkout with no ref:, which resolves to the default branch — not the PR being reviewed. The PR head is resolved separately (HEAD_SHA = gh pr view ... --json headRefOid).
Result:
- The review diff and comment anchors are correct — the engine fetches them from the GitHub compare API keyed on
HEAD_SHA(repos/{repo}/compare/...), independent of the checkout. - But the agent's explore tools (
read_paths/grep) read from the engine's_content_root(), which falls back to the primary checkout (= default branch on dispatch) whenREVIEW_CONTENT_ROOTis unset. So on manual dispatch, the agent explores default-branch code, not the PR head.
The automatic pull_request trigger (the normal case) is unaffected — its checkout already is the PR merge ref.
Why the engine expects this
Per reviewer_bot/run_review.py _content_root(), workflow_dispatch is meant to check the PR head into a separate directory and export its path as REVIEW_CONTENT_ROOT, while engine code keeps running from the trusted default-branch checkout. This is a deliberate security boundary: dispatch is exempt from the fork guard, so the PR's own code must not run/read from the primary checkout dir.
Fix
On the dispatch path, add a second checkout of HEAD_SHA into a separate dir and export REVIEW_CONTENT_ROOT pointing at it (mirroring the engine's documented pattern).
Severity
Low — only manual dispatch is affected, and only the explore-tool reads (diff/anchors are always correct). Deferred from #862 as a follow-up.
- 主要语言
- Python
- 星标
- 233
- 派生
- 152
- 平均合并
- 21 小时 5 分钟
- 30 天内合并 PR
- 10
贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
databricks/databricks-sql-python 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
-
难度 2/5 1-3 小时 新手友好度 76/100
-
难度 2/5 1-3 小时 新手友好度 78/100
-
难度 2/5 1-3 小时 新手友好度 72/100
-
engineer-bot
难度 2/5 1-3 小时 新手友好度 76/100
databricks/databricks-sql-python#860 · 3 条评论 ·
查看 databricks/databricks-sql-python 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
syfoud/Simulated_Scepter#172 ·
-
A cancelled tests run makes the coverage comment workflow fail and reports it as a red check on main 未关闭area: ci bug perceived difficulty: 3
难度 2/5 1-3 小时 新手友好度 78/100
Nitjsefnie-Harness-Commons/daedalus#921 · 1 条评论 ·
-
难度 2/5 1-3 小时 新手友好度 86/100
EleutherAI/lm-evaluation-harness#4207 ·
-
难度 1/5 1 小时以内 新手友好度 92/100
-
难度 2/5 1-3 小时 新手友好度 78/100
ClickHouse/clickhouse-connect#1057 ·