You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Filed by the director seat (#12708, summon #11, session session_01ShyhexkB2d1AeRZ85tgAAe) as the re-enable half the quarantine PR #14871 promised in its own body: "if that PR must not touch this test file, the director seat lands the one-line un-skip the moment the fix merges." The fix has merged — PR #14875 (Fixes #14832, merged 2026-09-03T11:34:02Z) — and it did not touch the test file, so this card is that un-skip. pm:queue · domain:cli · priority:p1: a maintainer-ruled gate weakening (ruling A on #14832, 2026-09-03, 「A,你现在就派发处理」) is not a resting state; restoring the gate is the closing act of that ruling.
What to restore, in packages/cli/test/run-dev-unbuilt-workspace.e2e.test.ts
PR #14871 (392f4108 on main) made exactly three edits; reverse all three, nothing else:
it.skip('gives up and exits instead of waiting forever', …) → it(…), and delete the quarantine note above it that cites ruling A.
The 'never-read' spawn in beforeAll (runCliAgainstDeadReader(REAL_COMMAND, dir, …, 'never-read', UNREAD_HARD_CAP_MS)) was commented out verbatim beside a restore note — un-comment it and delete the note.
⚠️ Re-derive from origin/main, not from #14871's diff alone: #14863 (#14716) edited the same file after the quarantine landed (the closed-read-end case is now judged by exit status, not a wall clock). Read the file as it is; the three edits above are the only ones this card makes. ⛔ bin/run-dev.js, bin/stderr-nonblocking.mjs, RUN_TIMEOUT_MS, UNREAD_HARD_CAP_MS and every other case stay untouched. ⛔ No timeout raised, nothing skipped, nothing retried.
Executable criterion
On the restored file, the case the mirror direction: a reader that is never coming back › gives up and exits instead of waiting forever runs (not skipped) and passes on a cold tsx cache repeatedly — the shape that reproduced the hang 27/30 before #14875 (per its PR body): run the file's suite under repetition with a private TMPDIR per run and report the observed pass count; every run must end code 2, signal null in seconds, never at the 180 s cap. The suite's check-test-completeness attestation must show the case counted as run.
Landing
Fixes this card; Part of #14832 in the body so the thread closes the loop. Tests-only ⇒ skip-changeset. Draft PR through the merge queue after the director seat's review; the Clause-② reading is no (test file only).
Refs: #14832 (the hang; ruling A recorded at 5521280392) · PR #14871 (the quarantine) · PR #14875 (the fix) · PR #14863 (#14716, the same file's later edit).
Filed by the director seat (#12708, summon #11, session
session_01ShyhexkB2d1AeRZ85tgAAe) as the re-enable half the quarantine PR #14871 promised in its own body: "if that PR must not touch this test file, the director seat lands the one-line un-skip the moment the fix merges." The fix has merged — PR #14875 (Fixes #14832, merged 2026-09-03T11:34:02Z) — and it did not touch the test file, so this card is that un-skip.pm:queue·domain:cli·priority:p1: a maintainer-ruled gate weakening (ruling A on #14832, 2026-09-03, 「A,你现在就派发处理」) is not a resting state; restoring the gate is the closing act of that ruling.What to restore, in
packages/cli/test/run-dev-unbuilt-workspace.e2e.test.tsPR #14871 (
392f4108onmain) made exactly three edits; reverse all three, nothing else:it.skip('gives up and exits instead of waiting forever', …)→it(…), and delete the quarantine note above it that cites ruling A.'never-read'spawn inbeforeAll(runCliAgainstDeadReader(REAL_COMMAND, dir, …, 'never-read', UNREAD_HARD_CAP_MS)) was commented out verbatim beside a restore note — un-comment it and delete the note.let unread!: Lifetime(the definite-assignment assertion added so the file stayed strict-clean while nothing assigned it) → back to the declaration form the file used before test(cli): quarantine the never-read-reader hang case behind #14832 (maintainer ruling A, 2026-09-03) #14871.origin/main, not from #14871's diff alone: #14863 (#14716) edited the same file after the quarantine landed (the closed-read-end case is now judged by exit status, not a wall clock). Read the file as it is; the three edits above are the only ones this card makes. ⛔bin/run-dev.js,bin/stderr-nonblocking.mjs,RUN_TIMEOUT_MS,UNREAD_HARD_CAP_MSand every other case stay untouched. ⛔ No timeout raised, nothing skipped, nothing retried.Executable criterion
On the restored file, the case
the mirror direction: a reader that is never coming back › gives up and exits instead of waiting foreverruns (not skipped) and passes on a coldtsxcache repeatedly — the shape that reproduced the hang 27/30 before #14875 (per its PR body): run the file's suite under repetition with a privateTMPDIRper run and report the observed pass count; every run must endcode 2, signal nullin seconds, never at the 180 s cap. The suite'scheck-test-completenessattestation must show the case counted as run.Landing
Fixesthis card;Part of #14832in the body so the thread closes the loop. Tests-only ⇒skip-changeset. Draft PR through the merge queue after the director seat's review; the Clause-② reading isno(test file only).Refs: #14832 (the hang; ruling A recorded at 5521280392) · PR #14871 (the quarantine) · PR #14875 (the fix) · PR #14863 (#14716, the same file's later edit).