test(ci): make web-shell visuals publish stub module-safe (#10736) - #10738
Merged
Conversation
The two end-to-end publish-step tests added in #10253 write an extensionless fake `gh` whose only module-system-dependent line uses `require('node:fs')`. Node resolves the module type of an extensionless script from the nearest ancestor package.json, so a stray `type: module` package.json anywhere above the temp root (shared CI hosts accumulate them) makes Node load the stub as an ES module. Every simulated `gh` call then fails, the publish step exits before writing the comment body the tests read, and both tests die on ENOENT — this was the main CI failure on 1804683, filed as #10736. Switch the stub to `process.getBuiltinModule('node:fs')`, which works in both module scopes, exactly as #10402 already did for the classify-release-notes stub.
Collaborator
Author
Autofix report for #10736 — Main CI failed: Qwen Code CI on 1804683DiagnosisIssue #10736 tracks a post-merge The failing commit is docs-only, so every pre-test step of the Root causeBoth tests were added by #10253 (merged 2026-09-01 11:12, hours before the failing run) and execute the real Node resolves the module type of an extensionless script from the nearest ancestor With every simulated This is the exact bug class #10402 already fixed in the sibling FixOne line in -require('node:fs').appendFileSync(process.env.CALL_LOG, args.join(' ') + '\n');
+process.getBuiltinModule('node:fs').appendFileSync(process.env.CALL_LOG, args.join(' ') + '\n');Commit: Mutation probe
The probe ran on a host where the hostile condition (a stray VerificationCommands actually run on the failing tree (
Not runnable in this sandbox (environment limitation, unrelated to the change): 中文说明#10736 自动修复报告 — 主分支 CI 失败：Qwen Code CI @ 1804683诊断#10736 跟踪的是一次合并后的 由于失败提交仅改动文档，我们在与失败提交完全一致的代码树上、按 CI 顺序本地复现了 根因两个失败测试均由 #10253 引入（2026-09-01 11:12 合并，距失败运行仅数小时），它们在带有假 Node 依据最近的祖先 由于每一次模拟的 这正是 #10402 已于 2026-08-28 在姊妹文件 修复
-require('node:fs').appendFileSync(process.env.CALL_LOG, args.join(' ') + '\n');
+process.getBuiltinModule('node:fs').appendFileSync(process.env.CALL_LOG, args.join(' ') + '\n');提交： 变异探针
探针运行在敌对条件真实存在的主机上（临时目录上方确有残留的 验证在失败代码树（
本沙箱无法执行（环境限制，与改动无关）： 🧠 Handled by Qwen Code · model/模型 |
github-actions
Bot
added
the
review/self-reported
Collaborator
|
|
Collaborator
|
Thanks for the fix! Template: complete — all required headings present, bilingual body ✓ Problem: observed, not theoretical. The post-merge Direction: in scope. Keeping the Size: 1 file, +1/−1, 0 production-logic lines (the only changed file is a Approach: minimal. The single line swaps Risk: no elevated signals — only a test stub changed, and the Stage 1e high-risk-path check (excluding test files) matched nothing. Moving on to code review. 🔍 中文说明感谢修复！ 模板： 完整——所有必需标题齐全，中英双语 ✓ 问题： 已观测到，非理论问题。合并后 方向： 在范围内。保持 规模： 1 个文件，+1/−1，0 行生产逻辑（唯一改动文件是 方案： 最小化。单行将 风险： 无升级信号——仅改动测试桩，Stage 1e 高风险路径检查（排除测试文件后）无命中。 进入代码审查 🔍 — Qwen Code · qwen3.8-max Reviewed at |
Collaborator
Code reviewIndependent proposal first: for an extensionless Findings: no blockers. No sequence diagram or changed-files table — this is a one-line, single-file change; both would be noise. CI test evidence (the PR's own CI, read via API — I do not run PR code here)The job that matters is Final CI results for
One row per check name (latest run); skipped checks omitted; failures sort first. / 每个检查名一行（取最新一次运行），省略 skipped，失败项排在最前。 What this evidence does and doesn't settle. Green CI on a clean runner shows the suite passes there, but the two previously-failing tests pass on a benign runner with or without this one line — the fix is only load-bearing under the hostile condition (a stray Sandboxed verification would settle the one open point:
中文说明代码审查先给出独立方案：对一个必须在任意模块类型下都能运行的无扩展名 结论：无阻塞项。 不附时序图或改动文件表——单文件单行改动，两者只会是噪音。 CI 测试证据（读取该 PR 自身的 CI， via API——此处不运行 PR 代码）关键任务是 该证据能/不能证明什么：在干净运行器上为绿说明套件在那里通过，但此前失败的两个测试在良性运行器上无论有无这一行都会通过——该修复仅在敌对条件（临时目录上方残留 沙盒验证可了结唯一悬而未决之点：
— Qwen Code · qwen3.8-max Reviewed at |
Collaborator
|
Confidence: 5/5 — clean across every stage; a one-line, precedent-aligned fix for a real, reproduced break on Stepping back: the problem is an observed post-merge CI failure on The one thing I could not independently confirm is the hostile-condition witness (revert → the two tests break again under a stray Verdict: approve. CI is still running on the reviewed commit, so approval is deferred until it lands green on 中文说明置信度：5/5 —— 各阶段均干净；一行、与先例一致的修复，针对 退一步看：问题是 唯一未能独立确认的是敌对条件见证（回退该行→在残留 结论：批准。由于被审提交上的 CI 仍在运行，批准推迟至其在 — Qwen Code · qwen3.8-max Reviewed at |
Contributor
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed. Suggestions are inline.
Not explored to full depth (tool budget reached): "agent 6b": none — no checks were cut short..
中文说明
已审查。 建议见行内评论。
未探索到全部深度（达到工具调用预算）："agent 6b"：none — no checks were cut short.。
— qwen3.8-max via Qwen Code /review (v0.22.3)
Collaborator
Author
|
🔀 Base updated: red check(s) [Test (ubuntu-latest, Node 22.x)] pass on current main — merged current main via update-branch; CI will re-run. 中文说明🔀 已更新 base：红色检查 [Test (ubuntu-latest, Node 22.x)] 在当前 main 上通过 —— 已通过 update-branch 合入当前 main，CI 将重新运行。 |
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Partially reviewed — gaps disclosed.
1 Suggestion-level finding(s) this review confirmed are already reported on this PR and are not repeated:
- R1-1 host-contingent regression guard for the gh stub's module-scope immunity — already reported (comment 3906901207)
Not reviewed: reverse audit — stopped before round 1 by the review time budget.
Deferred under the convergence posture (round 2, not a blocker) — recorded, not requested in this round:
.github/scripts/web-shell-visuals-publish.test.mjs:515 — [probe] D2-1 Fixes #10736 claim not backed by the cited run — its log shows the helper tests 507/507 green; the real failure was ChatEditor.test.tsx (fixed on main by #10729)
中文说明
仅完成部分审查，审查缺口已披露。
本轮确认的 1 条建议级发现已在 PR 上报告过，不再重复发布（列表见上方英文部分）。
未审查：反向审计——评审时间预算不足，未能开始第 1 轮。
收敛姿态下延后（第 2 轮，非阻断）——已记录，本轮不要求修改：共 1 条（原文未翻译，列表见上方英文部分）。
— qwen3.8-max via Qwen Code /review (v0.22.3)
Collaborator
Author
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 1/10). What changed, and what I pushed back on: · 已处理最新评审反馈（第 1/10 轮）。改动内容与我反驳保留之处如下： Round summary (PR #10738, issue #10736)One Suggestion-level inline finding this round; resolved in code after reproducing the claimed defect on the current code. No conflicts ( Feedback points and dispositions
Changed files
Verification
中文说明本轮总结（PR #10738，issue #10736）本轮共 1 条 Suggestion 级行内发现，已在先复现当前代码上的缺陷后于代码中解决。无冲突（ 反馈点及处置
变更文件
验证
Base-conflict check · 基分支冲突检查: no conflict with main. · 与 main 无冲突。 🧵 Resolved all 1 selected review thread(s). · 已关闭全部选中的 1 条评审线程。 Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. · 有空请复审；第 10 轮后本 bot 停止并将 PR 交给人工。 🧠 Handled by Qwen Code · model/模型 |
Collaborator
|
Periodic-review verification report — no Critical found (head What the PR does: the fake Verification (tmux-captured, scratch tree at head):
CI at head: 12 pass (incl. Not approving: no bot/maintainer approval on record at head (ci-bot's latest is COMMENTED). No equivalent prior feedback found; no other qqqys comment on this PR. 中文说明定期巡检验证报告 —— 未发现 Critical（head 该 PR 给端到端发布步骤测试用的 验证（tmux 捕获，head 源码树）：head 全绿 39/39；变异 A/B —— 仅把该行改回 — qqqys via Qwen Code periodic review |
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed — no blockers. Suggestions are inline.
中文说明
已审查——无阻断问题。 建议见行内评论。
— qwen3.8-max via Qwen Code /review (v0.22.3)
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 2/10). What changed, and what I pushed back on: · 已处理最新评审反馈（第 2/10 轮）。改动内容与我反驳保留之处如下： Autofix review-round summary — PR #10738 (issue #10736)Feedback points and decisions
Reproduction (source-blind verification)The finding's claim was reproduced on the pre-change head: with the pin line deleted, Changes (commit 9a720de)Test-only, single file
Mutation probes (each via
|
| Probe | Mutation | Result |
|---|---|---|
| Baseline | none | 40/40 pass |
| Pin witness | deleted the writeFileSync(join(bin, 'package.json'), …) pin line |
not ok 40 — the new witness test is the ONLY failure (39 pass / 1 fail); restored afterwards |
| Regression witness | reverted the default stub line to require('node:fs') |
not ok 38, not ok 39 — the two existing publish-step e2e tests still catch the original #10736 regression through the refactor; restored afterwards |
| Final | none | 40/40 pass |
The witness's failure reason was confirmed correct: without the pin, the require-based stub parses as CJS on this (CJS-ambient) host, the publish completes, and the body === null assertion fails.
Conflict notes
None — --conflict false, no base merge performed.
Verification
node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs— 40/40 pass, exit 0 (also the command CI'sHELPER_TESTSlane runs)npm run lint— passed (exit 0)npx prettier --check .github/scripts/web-shell-visuals-publish.test.mjs— passednpx eslint .github/scripts/web-shell-visuals-publish.test.mjs— passednpm run build— passed (exit 0)npm run typecheck— passed (exit 0)- Focused Vitest for touched packages — N/A (no package source touched; the change is a
.github/scriptshelper test exercised vianode --test) - Integration tests after
npm run bundle— N/A (behaviour exercised inside the node:test harness, not the bundled CLI)
中文说明
Autofix 审查轮次总结 — PR #10738（issue #10736）
反馈点与决定
- [rv:5083130173]（review 正文，COMMENTED —— "无阻断问题，建议见行内"）：信息性内容，无需处理。
- [rc:3908461047] R3-1（[Suggestion]，
.github/scripts/web-shell-visuals-publish.test.mjs:497）：本 PR 为gh桩新增的 ESM 作用域固定本身没有测试见证——删除writeFileSync(join(bin, 'package.json'), …)这行后整个套件依旧全绿，因此未来某次 fixture 重构可能悄悄删掉该固定行，使 Main CI failed: Qwen Code CI on 180468306c3a #10736 所描述的依赖主机的行为悄然回归。已在代码中解决（新增见证测试，并经变异探针验证）。 - [ic:5500330713]（@qqqys 定期巡检验证报告）：信息性内容——未发现 Critical，未要求任何修改，无需处理。
复现（来源盲验证）
该发现的论断已在修改前的 head 上复现：删除固定行后，node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs 依旧 39/39 全绿——没有任何测试观察该固定行。随后已恢复文件再开始实现。
修改内容（提交 9a720de）
纯测试、单文件 .github/scripts/web-shell-visuals-publish.test.mjs，按该发现建议的形态实现：
- 桩参数化：将
gh桩源码提升为模块级常量ghStubSource（逐字节原样搬移），并为runPublishStep增加ghStub选项，默认取该常量。 - 结果读取容错：当相应产物不存在时，
runPublishStep现在返回body: null、calls: ''、hostingStatus: null—— 这是必需的，因为启动即死的桩会让步骤停在机器人身份检查处，而该处在写出任何这些产物之前就以 0 退出。 - 新见证测试
'publish workflow fixture forces ESM scope onto the gh stub'：用 Main CI failed: Qwen Code CI on 180468306c3a #10736 之前的require('node:fs')桩行（通过replace从ghStubSource复现该行）运行发布步骤，然后断言发布未完成：步骤仍以 0 退出（工作流中的if [ -z "${BOT_LOGIN}" ]; then … exit 0; fi），runner/visuals-comment.md从未写出（body === null），且没有任何 gh 调用被记录（calls === ''）。按该发现的要求，断言落在缺失的产物上，而不是失败的退出码。
变异探针（均通过 node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs 执行）
| 探针 | 变异 | 结果 |
|---|---|---|
| 基线 | 无 | 40/40 通过 |
| 固定行见证 | 删除 writeFileSync(join(bin, 'package.json'), …) 固定行 |
not ok 40 —— 新见证测试是唯一失败项（39 过 / 1 败）；随后恢复 |
| 回归见证 | 将默认桩的那行改回 require('node:fs') |
not ok 38、not ok 39 —— 原有两个发布步骤端到端测试在重构后仍能捕获最初的 #10736 回归；随后恢复 |
| 最终 | 无 | 40/40 通过 |
见证测试的失败原因已确认正确：没有固定行时，基于 require 的桩在本机（CJS 环境作用域）被按 CJS 解析，发布正常完成，body === null 断言失败。
冲突说明
无 —— --conflict false，未做 base 合并。
验证
node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs—— 40/40 通过，退出码 0（也是 CIHELPER_TESTS流水线实际运行的命令）npm run lint—— 通过（退出码 0）npx prettier --check .github/scripts/web-shell-visuals-publish.test.mjs—— 通过npx eslint .github/scripts/web-shell-visuals-publish.test.mjs—— 通过npm run build—— 通过（退出码 0）npm run typecheck—— 通过（退出码 0）- 受影响包的聚焦 Vitest —— 不适用（未改动任何包源码；本修改是经由
node --test运行的.github/scripts辅助测试） npm run bundle后的集成测试 —— 不适用（行为在 node:test 测试装置内验证，不经捆绑 CLI）
Base-conflict check · 基分支冲突检查: no conflict with main. · 与 main 无冲突。
🧵 Resolved all 1 selected review thread(s). · 已关闭全部选中的 1 条评审线程。
Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. · 有空请复审；第 10 轮后本 bot 停止并将 PR 交给人工。
🧠 Handled by Qwen Code · model/模型 qwen3.8-max
Collaborator
|
Qwen Code review did not complete successfully. The review pipeline failed before a review could be posted. A transient error is retried automatically; if you are seeing this, retry with |
Collaborator
Author
|
🔀 Base updated: red check(s) [Post Coverage Comment (ubuntu-latest, 22.x)] pass on current main — merged current main via update-branch; CI will re-run. 中文说明🔀 已更新 base：红色检查 [Post Coverage Comment (ubuntu-latest, 22.x)] 在当前 main 上通过 —— 已通过 update-branch 合入当前 main，CI 将重新运行。 |
wenshao
pushed a commit
to wenshao/qwen-code
that referenced
this pull request
Sep 2, 2026
Collaborator
Maintainer verification report — built and run locallyVerdict: the change is good and I'd merge it — but Environment: Linux, Node 1. What actually reddened
|
| CI step | Result in that run |
|---|---|
34 — Run .github/scripts helper tests — the suite this PR touches |
# tests 507 # pass 507 # fail 0 ✅ |
36 — Run tests and generate reports |
❌ FAIL components/ChatEditor.test.tsx — ERROR: The symbol "language" has already been declared |
git show 180468306c:packages/web-shell/client/components/ChatEditor.test.tsx has language = 'en', listed twice in one destructuring pattern (a merge skew). esbuild refused to transform the file, vitest reported Test Files 1 failed | 247 passed, and the step exited 1.
That was fixed by #10729 (5282abea22, merged 2026-09-01T17:17:59Z) — eight minutes after this PR was opened (17:10:08Z). The stray-package.json condition this PR defends against never occurred on the CI host in that run.
Ask: drop Fixes #10736 (or restate it as "hardening found while investigating #10736"). As written, merging this closes #10736 crediting a fix that could not have fixed it, which will mislead anyone who later greps the issue for the ChatEditor regression.
2. Why the autofix lane was pointed at the wrong suite (not this PR's fault — worth a separate fix)
main-ci-failure-issue.yml fires on workflow_run: completed and fetches each failed job's log in a single shot with no retry. It ran 20 s after the job ended (16:21:32 → 16:21:52), GitHub's log endpoint was not ready yet, and the step logged ##[warning]Could not download the log of job 99922120921. The analyzer therefore ran with zero log files and fell back to the "failed before any test result was reported" body — which is what sent the agent to reproduce steps locally, where its own host's stray /tmp/package.json produced a completely different failure.
The extractor itself is fine. Re-fetched today, the log is 35 398 lines and extractFailingTests() returns exactly components/ChatEditor.test.tsx on the first try. A short retry/backoff (or a 30–60 s settle) around that gh api .../logs call would have produced the right issue and, presumably, the right autofix.
3. The change itself — verified, and it does more than the description says
I reproduced the hostile condition exactly (one stray {"type":"module"} package.json above TMPDIR) and ran the 2×2:
| arm | TMPDIR | tests | pass | fail |
|---|---|---|---|---|
BASE origin/main |
normal | 39 | 39 | 0 |
BASE origin/main |
hostile | 39 | 37 | 2 — ENOENT … /visuals-comment.md |
HEAD #10738 |
normal | 40 | 40 | 0 |
HEAD #10738 |
hostile | 40 | 40 | 0 |
Four mutation probes, all killed — every line of the diff is load-bearing:
| probe | mutation | result |
|---|---|---|
| M1 | process.getBuiltinModule('node:fs') → require('node:fs') |
2 fail — on a normal host, which is the real win |
| M2 | delete writeFileSync(join(bin,'package.json'), '{"type":"module"}') |
witness test 40 fails |
| M3 | drop the existsSync() guard on the body read |
witness test 40 fails |
| M4 | process.getBuiltinModule absent (simulates Node < 22.3) |
2 fail on HEAD; BASE passes 39/39 |
The important point M1 makes: before this PR the regression was only observable on a host that happened to carry a stray package.json; after it, the fixture pins ESM scope itself, so the same regression fails everywhere. That is a genuinely better test than the one-line #10402 treatment the description compares itself to — and it is exactly the part the description does not mention.
Other checks: prettier --check clean, eslint clean, no require( left in any .github/scripts stub after this lands, and the file already used this same {"type":"module"} fixture pin at line 240 for an unrelated sandbox, so the pattern is internally consistent. (My local full-HELPER_TESTS run shows 36 pre-existing failures, all in qwen-triage-workflow.test.mjs and identical on both arms — local environment, unrelated.)
4. Smaller points
- PR body is stale and understates the diff. It says "Expect 39/39" and "After:
# tests 507 … # pass 507"; the PR adds a 40th test, so it is 40/40 and 508/508. More importantly the body describes only the one-linegetBuiltinModuleswap, while thebin/package.jsonESM pin, theexistsSyncguards and the witness test are the bulk of+58/−29. Please fold those into the description — the pin is the best part of this PR and a reader of the body would never know it is there. - The pin narrows the tolerated Node range for this suite. Forcing ESM makes
process.getBuiltinModulemandatory, and it landed in Node 22.3.0, whilepackage.jsondeclaresengines.node: ">=22.0.0". Probe M4 shows the concrete effect: on a Node without it, HEAD fails 2 tests where BASE passes 39/39. CI resolves22.xto latest so this is theoretical today, and#10402already set the precedent — but if you want the floor honest,>=22.3.0is the one-line change. - Nit — diagnosability regressed slightly. With the guards, a genuine "the step never wrote the body" regression now fails as
Expected values to be strictly equal: null !== 'failure'. The old unguarded read failed withENOENT … /runner/visuals-comment.md, which named the missing artifact. Worth passingresult.stderr(or the path) as the assertion message so the next person does not have to re-derive it. - The witness's
assert.equal(result.status, 0)is correct, not a bug:web-shell-visuals-publish.yml:179deliberatelyexit 0s with a::warning::when the bot identity cannot be resolved. Worth knowing that a genuinely brokenghin real CI silently skips the visuals comment rather than failing the lane — pre-existing, out of scope here.
中文版报告
维护者本地验证报告
结论：改动本身是好的，建议合入；但 Fixes #10736 是错的，应当去掉。 本 PR 所修改的测试套件，在 issue 所引用的那次 CI 运行中是绿的；真正的失败是 ChatEditor.test.tsx 中重复的 language 绑定，而 #10729 已经修复了它。这里的加固仍然值得单独合入——它把一个依赖主机环境的潜在缺陷变成了在任何主机上都能被捕获的缺陷——所以这是一次来源归属的纠正，而不是要求关闭本 PR。
环境：Linux，Node v22.22.2。两个 worktree —— ARM=HEAD 为 PR head 3f8c2eccef，ARM=BASE 为 origin/main fde6565aae。
1. 究竟是什么染红了 main（阻塞项，但只针对 issue 链接）
我拉取了 issue 所引用那次运行的原始任务日志（33526417145，job 99922120921，提交 180468306c）：
| CI 步骤 | 该次运行中的结果 |
|---|---|
34 — Run .github/scripts helper tests —— 本 PR 修改的套件 |
# tests 507 # pass 507 # fail 0 ✅ |
36 — Run tests and generate reports |
❌ FAIL components/ChatEditor.test.tsx —— ERROR: The symbol "language" has already been declared |
git show 180468306c:packages/web-shell/client/components/ChatEditor.test.tsx 中，language = 'en', 在同一个解构模式里出现了两次（合并串味）。esbuild 拒绝转换该文件，vitest 报告 Test Files 1 failed | 247 passed，该步骤以 1 退出。
真正的修复是 #10729（5282abea22，合并于 2026-09-01T17:17:59Z）——比本 PR 创建时间（17:10:08Z）晚八分钟。本 PR 所防御的"残留 package.json"条件，在那次运行的 CI 主机上从未发生。
请求： 去掉 Fixes #10736（或改写为"排查 #10736 期间发现的加固"）。按现状合入会以一个不可能修复该问题的改动去关闭 #10736，日后有人检索该 issue 找 ChatEditor 回归时会被误导。
2. 自动修复通道为何被指向了错误的套件（不是本 PR 的问题，但值得单独修）
main-ci-failure-issue.yml 在 workflow_run: completed 时触发，单次拉取每个失败任务的日志，没有重试。它在任务结束后 20 秒运行（16:21:32 → 16:21:52），GitHub 的日志端点尚未就绪，该步骤记录了 ##[warning]Could not download the log of job 99922120921。于是分析器在零个日志文件的情况下运行，回退到了"failed before any test result was reported"的正文——正是这一点让 agent 转而在本地复现各步骤，而它自己主机上残留的 /tmp/package.json 制造出了一个完全不同的失败。
提取器本身没有问题。今天重新拉取，该日志有 35 398 行，extractFailingTests() 一次就返回了 components/ChatEditor.test.tsx。在那次 gh api .../logs 调用上加一小段重试/退避（或 30–60 秒的沉降等待），本可以产出正确的 issue，进而产出正确的自动修复。
3. 改动本身——已验证，且它做的比描述所说的更多
我精确复现了敌对条件（TMPDIR 之上一个残留的 {"type":"module"} package.json），跑了 2×2：
| 分支 | TMPDIR | tests | pass | fail |
|---|---|---|---|---|
BASE origin/main |
正常 | 39 | 39 | 0 |
BASE origin/main |
敌对 | 39 | 37 | 2 —— ENOENT … /visuals-comment.md |
HEAD #10738 |
正常 | 40 | 40 | 0 |
HEAD #10738 |
敌对 | 40 | 40 | 0 |
四个变异探针全部被杀死——diff 的每一行都是承重的：
| 探针 | 变异 | 结果 |
|---|---|---|
| M1 | process.getBuiltinModule('node:fs') → require('node:fs') |
2 个失败——在正常主机上，这才是真正的价值 |
| M2 | 删除 writeFileSync(join(bin,'package.json'), '{"type":"module"}') |
见证测试 40 失败 |
| M3 | 去掉正文读取上的 existsSync() 保护 |
见证测试 40 失败 |
| M4 | process.getBuiltinModule 缺失（模拟 Node < 22.3） |
HEAD 2 个失败；BASE 通过 39/39 |
M1 说明的关键点：在本 PR 之前，该回归只有在恰好带有残留 package.json 的主机上才可观测；在本 PR 之后，fixture 自己钉住了 ESM 作用域，同样的回归在任何地方都会失败。 这比描述中所对标的 #10402 那种一行修法要好得多——而这恰恰是描述完全没有提到的部分。
其他检查：prettier --check 通过，eslint 通过，合入后 .github/scripts 的任何桩脚本中都不再有 require(；并且该文件第 240 行早已为另一个无关沙箱使用了同样的 {"type":"module"} fixture 钉法，模式内部是一致的。（我本地跑完整 HELPER_TESTS 有 36 个既有失败，全部位于 qwen-triage-workflow.test.mjs，两个分支上完全相同——属本地环境问题，与本 PR 无关。）
4. 次要问题
- PR 描述过时且低估了 diff。 描述说"预期 39/39"以及"修复后：
# tests 507 … # pass 507"；本 PR 新增了第 40 个测试，因此应为 40/40 与 508/508。更重要的是，描述只讲了一行getBuiltinModule替换，而bin/package.json的 ESM 钉法、existsSync保护和见证测试才是+58/−29的主体。请把这些补进描述——那个钉法是本 PR 最好的部分，只读描述的人完全不会知道它存在。 - 该钉法收窄了本套件可容忍的 Node 范围。 强制 ESM 使
process.getBuiltinModule成为必需，而它是 Node 22.3.0 才引入的，package.json声明的却是engines.node: ">=22.0.0"。探针 M4 展示了具体影响：在没有该 API 的 Node 上，HEAD 失败 2 个测试，而 BASE 通过 39/39。CI 会把22.x解析为最新版，所以今天这只是理论问题，且#10402已开先例——但若要让下限诚实，>=22.3.0是一行的改动。 - 小瑕疵——可诊断性略有退化。 加了保护之后，真正的"步骤从未写出正文"回归现在会以
Expected values to be strictly equal: null !== 'failure'失败。而此前未加保护的读取会以ENOENT … /runner/visuals-comment.md失败，直接点名了缺失的产物。建议把result.stderr（或该路径）作为断言消息传入，免得下一个人重新推导。 - 见证测试中的
assert.equal(result.status, 0)是正确的，不是 bug：web-shell-visuals-publish.yml:179在无法解析机器人身份时会刻意以::warning::并exit 0。值得知道的是，真实 CI 中一个真正损坏的gh会静默跳过 visuals 评论而不是让该通道失败——这是既有行为，不在本 PR 范围内。
wenshao
approved these changes
Sep 2, 2026
wenshao
enabled auto-merge
September 2, 2026 02:15
qwen-code-ci-bot
previously requested changes
Sep 2, 2026
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Deferred under the convergence posture (round 4, not a blocker) — recorded, not requested in this round:
.github/scripts/web-shell-visuals-publish.test.mjs:521 — [probe] D4-1 fixture pins module scope only for the gh stub; uploader stub left under host-dependent scope inference (the #10736 mechanism)
Convergence: round 4 posted 2 inline comment(s), 2 of them reported for the first time; the previous round posted 1 (1 new). Findings keep coming back to the same files: .github/scripts/web-shell-visuals-publish.test.mjs (findings in round 3; 2 more now). The rate of new findings is not falling. A cluster that keeps producing siblings usually means the fixes are treating instances of a shared root cause — triaging that cause before the next round, or splitting an independent cluster into its own pull request, tends to end the loop faster than fixing them one at a time. Batching the remaining fixes and verifying them before the next push, or dropping this PR's reviews to --severity-floor critical, keeps the loop from re-deriving the same set. (Observation only — nothing was withheld from this review because of this observation.)
中文说明
收敛姿态下延后（第 4 轮，非阻断）——已记录，本轮不要求修改：共 1 条（原文未翻译，列表见上方英文部分）。
收敛情况：第 4 轮发布了 2 条行内评论，其中 2 条是首次提出；上一轮发布了 1 条（其中 1 条首次提出）。发现反复回到同一批文件：.github/scripts/web-shell-visuals-publish.test.mjs（第 3 轮已出过发现，本轮又有 2 条）。新发现的产出速度没有下降。一个不断再生兄弟发现的簇，通常意味着逐条修复只在处理同一根因的实例——先定位并处理该根因，或把独立的簇拆成单独的 PR，通常比逐条修复更快结束循环。把剩余修复攒成一批、验证后再推送，或将本 PR 的评审降到 --severity-floor critical，可以避免循环反复推导同一组发现。（仅为观察——本轮评审未因此扣留任何内容。）
— qwen3.8-max via Qwen Code /review (v0.22.3)
Comment on lines
+519
to
+520
| // Force ESM scope onto the extensionless `gh` stub so a `require()` | ||
| // regression in it is caught here on every host (#10736). |
Collaborator
There was a problem hiding this comment.
[Critical] R4-1: [certifies-falsely] [new-surface] The Fixes #10736 attribution in this PR's description is not supported by issue #10736's own incident. The cited run (33526417145, commit 180468306c) shows the .github/scripts helper suite this PR repairs passing 507/507; the run's actual failure was a duplicate language declaration in ChatEditor.test.tsx, already fixed on main by #10729. Merging with the Fixes trailer auto-closes the issue crediting this PR for a run this diff does not alter, cementing a misdiagnosis in the record. The code change itself is sound — the suite passes 40/40 at HEAD and deleting the pin turns exactly the new witness test red — the defect is solely the closing claim.
Concretely, in that run's job 99922120921 the step Run .github/scripts helper tests finished success (# tests 507 / # pass 507 / # fail 0), and the failing step is Run tests and generate reports: ChatEditor.test.tsx:425:4: ERROR: The symbol "language" has already been declared, Test Files 1 failed | 247 passed (248). The duplicate declaration exists at 180468306c and is fixed on main by 5282abea22 (#10729), which is an ancestor of main but not of 180468306c — the incident was already repaired by another PR. The description's narrative (two publish-workflow tests failing with ENOENT) matches a local simulation of a hypothesized hostile host, not the cited run.
Witness:
BASE (run 33526417145 job 99922120921): step "Run .github/scripts helper tests" -> success, "# tests 507 / # pass 507 / # fail 0"; failing step -> "Run tests and generate reports", "ChatEditor.test.tsx:425:4: ERROR: The symbol \"language\" has already been declared". git: "5282abea22 fix(web-shell): remove duplicate language declaration in ChatEditor.test.tsx (#10729)" IS ancestor of main, NOT ancestor of 180468306c; 180468306c:ChatEditor.test.tsx has "language = 'en'," at lines 422 and 425. PR body at 3f8c2eccef: "Fixes #10736".
Suggested fix: replace Fixes #10736 with a non-closing reference (e.g. "Discovered while investigating #10736; the run's actual failure was fixed by #10729"), and restate "Why it's needed" to say the cited run's log shows the helper suite green (507/507) with the real failure in ChatEditor.test.tsx, while this PR hardens the latent host-contingent stub bug the investigation reproduced. Land the code as-is.
中文说明
本 PR 描述中的 Fixes #10736 归属缺乏 issue #10736 事件本身的支撑。被引用的运行（33526417145，提交 180468306c）显示：本 PR 修复的 .github/scripts 辅助测试套件以 507/507 全部通过；该运行真正的失败点是 ChatEditor.test.tsx 中重复的 language 声明，且已由 #10729 在 main 上修复。带着 Fixes 尾注合并将自动关闭该 issue，并把这次运行归功于本 PR——而本 diff 并未改变该运行任何步骤的结果，从而在记录中固化一个误诊。代码改动本身是健全的——HEAD 上套件 40/40 通过，删除 pin 恰好使新增见证测试变红——缺陷仅在关闭声明。
具体而言，在该运行的作业 99922120921 中，Run .github/scripts helper tests 步骤为成功（# tests 507 / # pass 507 / # fail 0），失败步骤是 Run tests and generate reports：ChatEditor.test.tsx:425:4: ERROR: The symbol "language" has already been declared，Test Files 1 failed | 247 passed (248)。重复声明存在于 180468306c，并已由 5282abea22（#10729）在 main 上修复；该提交是 main 的祖先但不是 180468306c 的祖先——即该事件已由另一 PR 修复。描述中的叙述（两个发布工作流测试以 ENOENT 失败）对应的是对假想敌对主机的本地模拟，而非被引用的运行。
见证：
BASE (run 33526417145 job 99922120921): step "Run .github/scripts helper tests" -> success, "# tests 507 / # pass 507 / # fail 0"; failing step -> "Run tests and generate reports", "ChatEditor.test.tsx:425:4: ERROR: The symbol \"language\" has already been declared". git: "5282abea22 fix(web-shell): remove duplicate language declaration in ChatEditor.test.tsx (#10729)" IS ancestor of main, NOT ancestor of 180468306c; 180468306c:ChatEditor.test.tsx has "language = 'en'," at lines 422 and 425. PR body at 3f8c2eccef: "Fixes #10736".
建议修复：将 Fixes #10736 改为非关闭式引用（例如"在调查 #10736 时发现；该运行实际的失败已由 #10729 修复"），并重写"为什么需要"一节：说明被引用运行的日志显示辅助套件为绿色（507/507），真正的失败在 ChatEditor.test.tsx，而本 PR 加固的是调查过程中复现出的、依赖主机环境的潜在桩脚本缺陷。代码可按原样合并。
— qwen3.8-max via Qwen Code /review (v0.22.3)
Comment on lines
+572
to
+574
| hostingStatus: existsSync(hostingStatusPath) | ||
| ? readFileSync(hostingStatusPath, 'utf8').trim() | ||
| : null, |
Collaborator
There was a problem hiding this comment.
[Suggestion] R4-2: The new hostingStatus absent-artifact fallback (null) is exercised by the new witness test but never asserted — the one-line mutations : null -> : '' and : null -> : 'success' both survive the whole suite. The witness test constructs exactly the artifact-missing scenario and pins body === null and calls === '', but not hostingStatus; result.hostingStatus is asserted only at lines 968 ('failure') and 980 ('success') — both artifact-present paths. A future edit flipping this fallback to '' or a misleading 'success' ships green, the helper silently drifts from the null contract the diff's own comment states, and the first consumer to rely on it fails for the wrong reason.
Witness:
mutant null->'success' without assertion: "# tests 40 / # pass 40 / # fail 0" (survives); with added assert.equal(result.hostingStatus, null): "not ok 40 ... expected: ~ / actual: 'success'", "# pass 39 / # fail 1" (killed); fix alone on unmutated code: "# pass 40 / # fail 0".
Add to the witness test, next to the existing body/calls assertions:
assert.equal(result.hostingStatus, null); // hosting status was never writtenIf the added assertion is removed, the witness test must go red — confirm with the mutation: flip the fallback to : 'success', run the suite, and check that not ok 40 appears.
中文说明
新增的 hostingStatus 缺失产物回退值（null）会被新的见证测试执行到，但从未被断言——单行变异 : null -> : '' 与 : null -> : 'success' 均能在整个套件下存活。见证测试恰好构造了产物缺失的场景，并钉住了 body === null 与 calls === ''，却没有钉住 hostingStatus；result.hostingStatus 只在第 968 行（'failure'）与第 980 行（'success'）被断言——两者都是产物存在的路径。未来任何把该回退值改为 '' 或误导性的 'success' 的修改都会绿色通过，辅助函数将悄悄偏离 diff 自身注释所声明的 null 契约，第一个依赖该契约的使用者将以错误的原因失败。
见证：
mutant null->'success' without assertion: "# tests 40 / # pass 40 / # fail 0" (survives); with added assert.equal(result.hostingStatus, null): "not ok 40 ... expected: ~ / actual: 'success'", "# pass 39 / # fail 1" (killed); fix alone on unmutated code: "# pass 40 / # fail 0".
修复：在见证测试中、紧邻现有的 body/calls 断言处，添加：assert.equal(result.hostingStatus, null); // hosting status was never written。
若移除所添加的断言，见证测试必须变红——请以变异确认：将回退值改为 : 'success'，运行套件，确认出现 not ok 40。
— qwen3.8-max via Qwen Code /review (v0.22.3)
qqqys
approved these changes
Sep 2, 2026
qqqys
left a comment
qqqys
left a comment
Collaborator
There was a problem hiding this comment.
APPROVE — test-only,独立验证 @ 3f8c2ecc
Reviewed at head 3f8c2eccef044041d1765673e346ab2560c31125.单文件测试改动(.github/scripts/web-shell-visuals-publish.test.mjs,+58/−29),通读完整 patch:
- 修复本体合理: 扩展名less 的
ghstub 改用process.getBuiltinModule('node:fs'),消除对宿主隐式 CJS scope 的依赖(#10736 的成因);fixture 侧以bin/package.json {"type":"module"}把 stub 钉进 ESM scope,使require()回归在所有宿主上可观测,不再是 host-contingent。 - 见证测试自洽: witness 以
require版 stub 运行整步,断言 status 0(与 workflow 在 BOT_LOGIN 为空时exit 0的行为一致,注释引用了具体分支)且 body/calls 缺失——删除 pin 即打红,满足变异敏感性;产物读取改为可选(existsSync+ null/'')只服务于该死亡-stub 形态,原有四个断言测试路径不变。 - 历史项: 本 PR 无 CHANGES_REQUESTED、无 Critical;两条行内 Suggestion(stub 免疫性 host-contingent;pin 无见证)分别由
6f29856c与9a720de7处理。维护者 01:46 的验证报告给出「the change is good and I'd merge it」并保留一条描述层意见(Fixes #10736应去掉),随后本人于本 head 提交 APPROVE——属 PR 元数据范畴,不影响代码结论。 - CI(非门禁): 本 head 14 success / 61 路径性 skip / 3 在途,无失败,无本 PR 引入的阻塞性缺陷。
✅ APPROVE — 当前扫描无可证明 Critical,历史无站立项,批准。
— 衍星 · read-only PR review (posted as qqqys)
Collaborator
|
Released in v0.23.0. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this PR does
Makes the fake
ghstub used by the two end-to-end web-shell visuals publish-step tests independent of Node's module-type resolution. The stub is an extensionless script; its only module-system-dependent line now usesprocess.getBuiltinModule('node:fs'), which works whether Node loads the file as CommonJS or as an ES module. This is the same one-line treatment #10402 already applied to the sibling release-classifier stub.Why it's needed
The post-merge
Qwen Code CIrun on commit180468306cfailed onmainbefore any unit-test result was reported (issue #10736). Reproducing the test job's pre-test steps in order locally showed the break was the.github/scriptshelper suite: the two publish-workflow tests added by #10253 failed withENOENTon the comment body file. Node resolves the module type of an extensionless script from the nearest ancestorpackage.json, so a straytype: modulepackage.json anywhere above the temp root — which shared CI hosts accumulate — made Node load the stub as an ES module whererequireis not defined. Every simulatedghcall then failed, the publish step exited before writing the comment body the tests read, and both tests died.Reviewer Test Plan
How to verify
Run the helper suite that failed on
main:node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs. Expect 39/39 passing. On a host that carries a straytype: modulepackage.json above the temp directory (the hostile condition that reddenedmain), the pre-fix code fails the twopublish workflow ...tests withENOENT .../visuals-comment.md; with this change they pass in both module-type environments. A mutation probe confirmed the witness: reverting the one line re-breaks exactly those two tests.Evidence (Before & After)
N/A — CI test-infrastructure change, no user-visible behavior. Before:
# tests 507, # pass 505, # fail 2for the HELPER_TESTS step. After:# tests 507, # pass 507, # fail 0.Tested on
Environment (optional)
Node 22.x, direct
node --testruns of the.github/scriptshelper suites;process.getBuiltinModulerequires Node ≥ 22.3, matching the repository's>=22floor and already relied on by the sibling stub.Risk & Scope
package.jsonresidues on shared hosts are not cleaned up here — the lane only needs the tests to tolerate them, as test(ci): make release classifier stub module-safe #10402 established.shellcheckandyamllintwere not runnable in the local sandbox (missingxz/pip3); neither covers a.mjstest file.Linked Issues
Fixes #10736
中文说明
本 PR 做了什么
让 web-shell 可视化发布步骤两个端到端测试所使用的假
gh桩脚本不再依赖 Node 的模块类型推断。该桩是一个无扩展名脚本；其唯一依赖模块系统的行现在改用process.getBuiltinModule('node:fs')——无论 Node 将文件按 CommonJS 还是 ES 模块加载均可工作。这与 #10402 已经对姊妹的发布说明分类器桩脚本采用的一行修法完全一致。为什么需要
提交
180468306c合并后的Qwen Code CI运行在main上、尚未报告任何单测结果之前失败（issue #10736）。按顺序在本地复现 test 任务的测试前步骤后，断点定位在.github/scripts辅助测试套件：#10253 新增的两个发布工作流测试以ENOENT（缺少评论正文文件）失败。Node 依据最近的祖先package.json判定无扩展名脚本的模块类型，因此只要临时目录上方存在任何残留的type: modulepackage.json——共享 CI 主机上会积累此类残留——Node 就会把该桩按 ES 模块加载，其中require未定义。于是每一次模拟的gh调用都失败，发布步骤在写出测试要读取的评论正文之前提前退出，两个测试随即失败。审阅者测试计划
如何验证
运行在
main上失败的辅助测试套件：node --test --test-concurrency=1 .github/scripts/web-shell-visuals-publish.test.mjs，预期 39/39 通过。在临时目录上方存在残留type: modulepackage.json 的主机（即染红main的敌对条件）上，修复前的代码会以ENOENT .../visuals-comment.md使两个publish workflow ...测试失败；本改动使它们在两种模块类型环境下均通过。变异探针确认了见证：回退这一行会令恰好这两个测试再次失败。前后证据
N/A —— CI 测试基础设施改动，无用户可见行为。修复前：HELPER_TESTS 步骤
# tests 507, # pass 505, # fail 2。修复后：# tests 507, # pass 507, # fail 0。测试平台
环境（可选）
Node 22.x，直接以
node --test运行.github/scripts辅助套件；process.getBuiltinModule需要 Node ≥ 22.3，与仓库>=22的下限一致，且姊妹桩脚本已在使用。风险与范围
package.json不在本 PR 清理——该测试通道只需要测试能够容忍它们，test(ci): make release classifier stub module-safe #10402 已确立此原则。本地沙箱缺少xz/pip3，shellcheck与yamllint无法运行；两者均不覆盖.mjs测试文件。关联 Issue
Fixes #10736