test budget (domain:devx): 2 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23185

Activity

1. objectstack-fleet

   added

   toolingFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surfaceFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surface

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   domain:devxDomain lane — seat card indexed by label:pm:seatDomain lane — seat card indexed by label:pm:seat

   priority:p3

   area:devpathThe road — create, dev, verify, publish/install, connect an agent, iterateThe road — create, dev, verify, publish/install, connect an agent, iterate

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   and removed

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Oct 13, 2026
2. objectstack-fleet

   assigned

   zhuangjianguo

   on Oct 13, 2026
3. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetBot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 11 · domain:devx seat 1 · 2026-10-13T12:00Z · triage-graded (priority:p3), part of #23151 under ruling E on #22949
   Session: session_012zwsLMxQBwHUYJgPyrpEvx
   Account: zhuangjianguo (the seat's linked user as GET /user answers it; the card's assignee)
   Branch: claude/issue-23185-showcase-slow-tests
   Worktree: objectstack-issue-23185
   Domain: domain:devx
   Seat: domain:devx#1
   File surface: examples/app-showcase/test/vitest-console-teardown-race.test.ts and examples/app-showcase/test/approval-resume-relation-expand.test.ts, and what they wait on inside examples/app-showcase if the measured cause is there. ⛔ No deletion without a named duplicate, ⛔ no timeout raise, ⛔ no move to a slower or skipped project (the card's own bans)
   Container & model: M, mode:subagent, model: opus (the default judgment tier)
   Clause-②: no (tests)
   Responsibility: two showcase test files run 20.8 s and 10.2 s against a 10 s budget | no platform path covers it: the budget rule (#22998) names the owning lane | the hourly full battery pays the wall on every run
   Thread-read: none
   Serial constraints cleared: no open PR touches either test file (all open PRs' file lists read at this act).
4. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetBot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   os-dev-report
   {
   "issue": 23185,
   "status": "done",
   "branch": "claude/issue-23185-showcase-slow-tests",
   "pr": "#23273",
   "session": "session_012zwsLMxQBwHUYJgPyrpEvx — subagent run, so the parent's harness-stamped id (the Claude-Session trailer on every commit of the branch)",
   "premise_still_valid": true,
   "summary": "Both rows of the card, one PR for the package, with Fixes #23185. Both files now replace their clocks with sync points. (1) vitest-console-teardown-race: measured that the race has TWO vitest windows, (a) the worker's teardown sweep finding a pending console RPC and (b) the main process grading the worker's late rejection report before it arrives. A new harness reporter, test/fixtures/late-console-teardown/hold-console-acks.reporter.ts, holds (a) open deterministically: console log N is acked only when N+1 arrives, so the newest RPC is always pending at the sweep. It reads (b) instead of racing it: at child exit it writes the delivered rejection count beside the graded one to a verdict file passed via vitest's --outputFile option. Each leg now runs ONCE, where before it was up to 8 ablation attempts plus 4 guarded repetitions. The skip path is gone, and the premise probe gains a 6th marker (the cli-api console RPC handler awaits the reporters) with a matching control case. The fixture's first tick is now synchronous. (2) approval-resume-relation-expand: the stays-quiet watch ends on a settle point (the stranded run is failed and every sys_notification_delivery row is final) instead of waiting out the 5 s window, and asserts endedBy=settled. The fixed sleep(200) is gone, because decide() awaits the resume. Three clock-free contract cases were added. File walls are 2.2-3.2 s and 2.1-2.5 s even beside four concurrent vitest loops, against 6.3/7.3 s locally before and 20.8/10.2 s on CI. Every touched file is inside examples/app-showcase (the two tests, the fixture dir, and types/node-shim.d.ts gaining only the members imported). The PR body carries 'Clause-②: no' as the dispatch spelled it. The claim's '(tests)' suffix is not one of the declared arms, so it was not copied.",
   "tests": "HEAD 5ea05a3 (the base was afbf960; main merged in, touching no file of this package). (a) cd examples/app-showcase && vitest run --maxWorkers=2 over the whole package: 'Test Files 33 passed (33) / Tests 412 passed (412)', 0 skipped. The box was loaded by the concurrent gate battery, and the two files read 2.95 s and 2.55 s in that run. (b) pnpm --filter @objectstack/example-showcase typecheck: exit 0, script name echoed; tsc --listFiles counts all 5 touched files. (c) The 39 families derived by node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 5ea05a3 were run, and --ran with recorded exit codes reads '39 derived, 38 run, 1 NOT-MEASURED, 0 UNRUN'. NOT MEASURED: pnpm check:dual-build-cjs-loads, reason: exit 3 PREREQUISITE NOT MET after its self-test passed (93 cases), because 8 packages outside the showcase closure (studio, client-react, embedder-openai, knowledge-memory, knowledge-ragflow, organizations, service-cluster-redis, service-knowledge) have no dist/ here and this diff changes no dist; CI reads it. Extra: node scripts/check-console-intercept-disarm.mjs OK (74 packages). (d) Lint as a proven narrowing: eslint --no-inline-config --format json over the 5 touched files reports 5 files, 0 errors, 0 warnings. eslint.config.mjs enables no type-aware linting (no parserOptions.project, no projectService), so untouched files' verdicts cannot move. (e) Before, at afbf960: the teardown pin took 6.34 s (ABLATION 2126 ms, GUARDED 4203 ms) and the approval file 7.30 s (its pre-fix case 5213 ms). After, at 53ed69f: the teardown pin took 2.2-2.5 s idle (5 runs) and 2.2-3.2 s beside four concurrent vitest loops (10 runs), green 15/15; the approval file took 2.19-2.30 s idle (3 runs) and 2.13-2.48 s under the same load (5 runs). (f) Ablations, all via node scripts/ablation-replace.mjs in wrap mode. Each showed anchor 1 to 0 and a changed blob; each restore showed the blob equal to HEAD and an empty git diff HEAD. Observed direction is red everywhere. A1, deleting disableConsoleIntercept: true from vitest.config.ts: GUARDED red 10/10 (5 idle, 5 loaded), verdict delivered=1. A2, the hold reporter returning Promise.resolve() (no hold), 12 loaded runs: ABLATION red 11/12, every one on the 'instrument' branch, never a skip. B1, the watch verdict read replaced by the pre-settle sample: 3 contract cases red. B2, the call site settling on completed instead of failed: the pre-fix case red on endedBy after paying the 5 s window (file 7.23 s). B3, inboxSettled dropping its outbox half: the 'both halves' contract red. (g) Prototype race measurements (a .mjs prototype of the same hold, one child per attempt): defaults without the hold reproduced 20/20 idle and 4/20 under 4 concurrent vitest loops; with the hold, rejections were 40/40 idle and 40/40 loaded (32/40 graded in time, which is window b); the app config with the hold gave 0/10; the guard re-armed via CLI with the hold gave 20/20. No build/dist ablation applies: the tests run source.",
   "mcp_calls": "0 — no MCP GitHub tool called",
   "api_writes": "3 — each a fleet-write relay stroke (POST /repos/objectstack-ai/objectstack/dispatches, executed as objectstack-fleet[bot]): pr_create (POST /pulls draft, then POST /issues/23273/assignees zhuangjianguo; body read back byte-identical), labels_add skip-changeset (POST /issues/23273/labels; read back: size/l, tests, skip-changeset), and this report (POST /issues/23185/comments via post-stamped.mjs). 4 git pushes, which are not REST.",
   "open_questions": [],
   "out_of_scope_findings": [
   "carrier: none (承接者:无) · noted, not filed — vitest's own window (b): a teardown rejection the worker reports after the pool settles is recorded but never graded (exit 0). This is upstream behaviour outside our contract, moot behind disableConsoleIntercept: true. It is in PR #23273's Acceptance notes.",
   "carrier: none (承接者:无) · noted, not filed — examples/app-showcase/vitest.config.ts's disarm docblock and the scripts/check-console-intercept-disarm.mjs header still describe the race window as a single load-width span inside rpcDone(). The measured second window refines but does not contradict either prescription. It is in PR #23273's Acceptance notes."
   ]
   }
5. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetBot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Review: ACCEPT · PR #23273 (draft, head 5ea05a3201) · domain:devx seat 1 · session_012zwsLMxQBwHUYJgPyrpEvx · 2026-10-13T13:38Z

   Checked on GitHub, not on the report's say-so.

   • Shape: five files, all inside examples/app-showcase (+592 / −204):
     • the two test files;
     • a new fixture reporter test/fixtures/late-console-teardown/hold-console-acks.reporter.ts;
     • the fixture test;
     • types/node-shim.d.ts, which only gains the members it imports.
   • Fixes #23185. Labels size/l, tests and skip-changeset, which is right for tests that ship in no dist. Not governed (0 of 5 paths), and no contract face is touched.
   • Against the card's Done-when and its bans:
     • Both files are under 10 s: 2.2–3.2 s and 2.1–2.5 s beside four concurrent vitest loops, against 20.8 / 10.2 s on CI. The fix is to the tests and what they wait on: clocks are replaced by sync points.
     • No deletion, no timeout raise, no move to a slower or skipped project. The skip path is removed.
     • One PR covers the package, which is smaller than one per file. It closes both rows.
   • Detection power is kept and proven. The dispatch made this the condition for the race test.
     • The reporter holds vitest's teardown window (a) deterministically: log N is acked only when N+1 arrives. Window (b) is read rather than raced, through a verdict file via --outputFile.
     • Ablation A1 removes disableConsoleIntercept: true, and the guard goes red 10/10, both idle and loaded.
     • Ablation A2 removes the hold, and the instrument goes red 11/12, never a skip.
     • The premise probe gains a sixth marker with a control.
     • In the approval file, the stays-quiet watch ends on a settle point and asserts endedBy=settled. Ablations B1, B2 and B3 each go red on the case they should.
   • Tests:
     • The whole package: 33 files and 412 tests pass, 0 skipped. Typecheck exit 0.
     • Gates: 39 derived, 38 run, 1 NOT-MEASURED (check:dual-build-cjs-loads, because packages outside this closure have no dist here; CI reads it).
     • eslint on the five files reports 0 errors.
   • CI on 5ea05a3201: 28 green, 10 skipped, 3 running, none red. The seat arms auto-merge.
   • Pin: the next shard-timings refresh PR no longer lists either file. That is this seat's refresh lane under ruling E.
   • Noted in the PR, not filed: the disarm docblock in vitest.config.ts and the check-console-intercept-disarm.mjs header still describe one race window, where two were measured. That refines them and does not contradict either.
6. objectstack-fleet

   closed this as completedin #23273on Oct 13, 2026
7. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetBot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Landed: PR #23273 → 6c181625f9 (Fixes #23185) · domain:devx seat 1 · session_012zwsLMxQBwHUYJgPyrpEvx · 2026-10-13T14:09Z

   • PR test(showcase): bring the teardown-race and approval-resume test files under the 10 s budget #23273 merged through the queue. On main, 5/5 files equal the PR head 5ea05a3201.
   • The card closed completed on the merge, and pm:dispatched is removed with this record.
   • What holds on main:
     • vitest-console-teardown-race.test.ts and approval-resume-relation-expand.test.ts replace their clocks with sync points.
     • The hold reporter keeps teardown window (a) open deterministically, and window (b) is read through a verdict file.
     • Both files measured 2–3 s against 20.8 / 10.2 s on CI.
     • Detection power is proven by ablation: removing the disarm reddens the guard 10/10.
   • Pin, open: the next shard-timings refresh PR no longer lists either file. This seat reads it under ruling E when that refresh lands.
8. objectstack-fleet

   removed

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   on Oct 13, 2026