Skip to content

Diagnose release-publish fixture timeouts in full unit validation #4471

Open

Description

@thymikee

Purpose

Diagnose release-publish fixture timeouts in full unit validation. Observed during corrections for now-merged #4432. Follow-up #4473 has a newer passing exact-head receipt; this issue retains the historical failure diagnosis under coordinator ownership. No worker dispatch is requested by this issue.

Evidence

At 39f2c75ea6be4e56a0563e55fa00af69df48ba95, pnpm check:affected --run --head 39f2c75ea6be4e56a0563e55fa00af69df48ba95 failed in scripts/__tests__/release-publish.test.ts:89, “publishes core after every plugin”: 5000ms test timeout. Receipt run-fpQgOY: exactCommit=true, checkoutChanged=false, clean source; 19,080 Vitest passed, one failed, three skipped. Node unit and later selected checks did not run. Format/lint/typecheck/static guards/build/package/integration/related stages passed.

The three-file follow-up restricts a test-kit type, adds its regression, and changes the daemon-fixture repository file scan. Both scripts/release-publish.mjs and its test are unchanged from prior head 18fad61f628ce7744b672b880d94a8da5d3d6002.

At that prior head, earlier receipt run-Z5Avmu failed another case in the same fixture (:113, large publish output). A later explicitly requested clean exact-head gate, run-7tAruR, passed all 42 runnable checks, including 19,080 Vitest/76 Node tests. That does not establish a cause for either failed receipt, or a matching base failure at line89.

One bounded phase experiment used the same command on candidate and prior head, in sequence in an isolated checkout after frozen install/build:

pnpm exec vitest run --project unit-core scripts/__tests__/release-publish.test.ts --reporter=verbose

A temporary NODE_OPTIONS preload recorded child-process start/settlement and process entry/exit without changing source, assertions, timeouts, worker policy or signals. Both runs passed five tests. In the line89 case, candidate release process2695.72ms, summed child calls2691.85ms; base3009.87ms, summed child calls3000.52ms. Each launched21 commands (five fake tar,16 fake npm); slowest command222.83/234.16ms. Instrumentation adds overhead. No actual failing phase was captured, and these passing measurements do not prove host contention or a fixture invariant defect.

Completion contract

Capture phases on the next actual failing full-suite occurrence using the existing command, default workers and budgets. Identify which startup, command, stream/close or teardown phase consumed the budget. Do not loop broad retries merely to obtain green.

If evidence identifies a construction or lifetime defect, fix its owning fixture or release boundary and prove detection before/after. Preserve all five contracts: plugin-failure aggregation/core suppression, publish order, skipping published versions, large-output streaming, and signal classification. Do not increase timeouts, add retries/slow pins, alter isolation/worker policy, skip/delete tests, weaken assertions or change the full-unit fallback.

Keep failed receipts failed until new named-commit evidence exists. No real npm publication or release dispatch. Raw phase traces, commands and receipts are retained in the maintainer ledger under runs/2026-11-12-pr-comments/4432-companion-review/; publish relevant excerpts with any correction so local paths are not the only evidence.

Activity

  1. apex-sentinel-staging commented on Nov 12, 2026

    @apex-sentinel-staging

    Apex Sentinel · Issue review

    Suggested label: bug (does not look actionable)

    The issue reports an intermittent 5000ms timeout in an existing release-publish test, which is a defect rather than a request for new behavior. However, the cited failing receipt is historical, the newer exact-head receipt passed, and the issue explicitly says no worker dispatch is requested and diagnosis depends on a future actual failing full-suite run, so there is little maintainer action to take now.

    Required sections

    • Steps to reproduce (missing)
    • Expected and actual behavior (missing)
    • Version or environment (missing)
  2. thymikee commented on Nov 12, 2026

    @thymikee
    MemberAuthor

    Current state: #4432 merged during the review-fix work. The two fixes are published in #4473 at 8e5db839694a17f7d11d6b558fb4b2610f3940b7, based on current main 59c52465. Range-diff preserves both patches unchanged. The normal selected exact-head gate for that new follow-up passed: receipt run-OHiTdY, exactCommit=true, checkoutChanged=false, eight runnable checks passed, seven CI-owned skips, no failures/unrun checks. Full unit passed 19,480 Vitest/76 Node tests (three/six skipped).

    That new-head validation does not relabel run-fpQgOY or run-Z5Avmu, and it does not explain either timeout. The phase experiment on pre-rebase candidate 39f2 and base 18fad passed five tests each; no failing phase was captured. No fixture, timeout, worker policy, assertions, retries or full-unit fallback changed. Diagnosis remains open for phase capture at a real failing occurrence; no additional broad retry loop is planned.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions