LearnNewsExamplesConnect
Frontmatter
assigneesneo-opus-grace
createdAtOct 26, 2026, 4:45 PM
updatedAtOct 26, 2026, 6:54 PM
githubUrlhttps://github.com/neomjs/neo/issues/20128
authorneo-opus-grace
commentsCount3
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]
closedAtOct 26, 2026, 6:54 PM

Sibling-holds e2e finds no Audit header when its sampler starts

neo-opus-grace
neo-opus-grace commented on Oct 26, 2026, 4:45 PM

Context

test/playwright/e2e/workstation/WorkstationTabDragSiblingHoldsNL.spec.mjs (#20074, for #20072) has failed twice with the same symptom:

  • 2026-10-23: Ada, while qualifying #20083, ran the grid and workstation NL tier: "a sampler race in WorkstationTabDragSiblingHoldsNL, 3/3 head + 3/3 base in isolation".
  • 2026-10-26: on the contributor PR #20119, Brain E2E shard 2/10 (run 40277307605, job 121458090653) failed at :59 with page.evaluate: TypeError: Cannot read properties of null (reading 'getBoundingClientRect'). The identical spec blob affff65e passed on dev 056f5493 with the same Brain pin, so #20119's diff is not implicated.

The Problem

Before the drag, the spec resolves the sibling's header id through getTabChromeIdentity('right-top-tabs') and the element is found (page.locator('#' + chrome.buttons.audit).boundingBox() succeeds). It then sets a 4× CPU throttle, starts dragging Metrics, waits for .neo-tab-header-toolbar.neo-is-dragging, moves 300 px below the strip, and only then looks the sibling up again inside page.evaluate (document.getElementById(id)). On a failing run that lookup returns null, so the pre-drag id has no element in the DOM at that moment.

Two readings remain, and they lead to different fixes:

  • (a) A product defect. Under load, the Audit header leaves the DOM or is re-rendered at drag start, so a user would see the sibling tab blink. This is the kind of sibling disruption the spec exists to forbid.
  • (b) A test artifact. The sibling persists, but its element identity changes between the pre-drag resolution and the sampler.

The Architectural Reality

  • The spec reads the chrome identity through the Neural Link (app.callMethod(workspaceId, 'getTabChromeIdentity', [NODE])) and the DOM through page.evaluate. The sampler captures header once and samples getBoundingClientRect().x every animation frame.
  • The disappearance happens inside the dock's tab-drag start path under throttle. Which component re-renders the strip there is not established yet; that is the first deliverable below.

The Fix

  1. Discriminate first. Instrument the sampler to re-resolve the sibling every frame and, on a frame where the id is absent, record the strip's header-button ids and classes. Reproduce under the spec's 4× throttle (--repeat-each) on dev.
  2. For (a): fix the drag-start path so the sibling header keeps its element, and make the spec fail on any frame where the sibling is absent.
  3. For (b): the spec resolves the sibling by its stable identity on every frame; no product change.

Acceptance Criteria

  • AC-1: a reproduction receipt names which case occurs: run count, null-frame count, and what the strip held on a null frame.
  • AC-2: the fix for that case lands; for (a), a red-first spec arm shows the absent sibling on dev.
  • AC-3: the spec treats an absent sibling as a failure rather than a TypeError, and passes 20/20 under the 4× throttle at the fix head.

Out of Scope

  • #20090 (DockMotionNL's moved-pane read), a separate dock e2e timing defect.
  • Hosted-CI scheduling of the NL tier (#19655).

Related

#20072 · #20074 · #20083 · #20119

Sweeps (2026-10-26T15:44Z): live latest-open, the 20 newest neomjs/neo issues, no equivalent (#20090 is adjacent but a different spec); keyword SiblingHolds / getTabChromeIdentity, none open; A2A, latest 30 messages, no claim on this spec; own-assignment, #20114, #19919 and #15000, none overlapping.

Origin Session ID: 9853627c-2129-4d41-b905-0c74bec35a48 Retrieval Hint: "WorkstationTabDragSiblingHoldsNL sampler null getBoundingClientRect Audit header drag start CPU throttle"

tobiu closed this issue on Oct 26, 2026, 6:54 PM
tobiu referenced in commit a62f106 - "fix(tab): a press on a tab holds the focus-gated reveal until it clicks or its drag ends (#20128) (#20135) on Oct 26, 2026, 6:54 PM