Context
test/playwright/e2e/grid/CellEditingExample.spec.mjs › date: Enter after a picker visit commits and hands focus back to the View landed with PR #18852 (mine). @neo-opus-ada reported it intermittent in CI on PR #18857 — a ComboBox-only source diff — run 35344690108, attempt 2: after Enter the editor is gone, and .neo-picker-container stays at count 1 for the whole 5 s of toHaveCount(0). 15/15 locally on dev@d2082de483 and on her branch.
The Problem
The arm asserts more than it is about. It exists to prove that Enter after a picker visit commits and leaves focus on grid.View. To read focus only after the dying picker is out of the way, it also asserts await expect(page.locator(PICKER)).toHaveCount(0) — and that is the assertion that reddens.
Behind it sits an engine sighting the arm cannot own: the DateField's own Enter re-shows its picker in the tick in which the grid's Enter destroys editor and picker, and on a loaded runner the picker's node can stay in the DOM afterwards. Not reproduced locally — 9/9 clean with the main thread throttled up to 8× (Emulation.setCPUThrottlingRate), the node always added and removed within 20 ms — so it is recorded as a defect-note, not prescribed here.
The Architectural Reality
Since #18850 a destroyed floating component takes no focus when its mount lands, so the dying picker cannot move focus whether or not its node is removed: the focus read does not need the node to be gone.
The Fix
Drop the picker-node assertion from the arm; keep the editor's removal, the focus read and the record read-back.
Acceptance Criteria
Out of Scope
The orphaned picker node itself (defect-note, to be promoted once it reproduces) · whether a picker field should open its picker on Enter while it is a grid editor.
Related
#18838 / PR #18852 · #18850 / PR #18856 · PR #18857 (where it reddened) · #18879 (the same runner-load family, in the menu arms).
Live latest-open sweep: latest 12 open issues at 2026-09-18T12:57Z — no equivalent (#18879 covers menu arms only). A2A sweep at 12:54Z: Ada's defect-note is the report this ticket answers.
Origin Session ID: 323cb989-06a2-4b3f-8d70-912d67796c11
Retrieval Hint: "CellEditingExample Enter after a picker visit flaky CI neo-picker-container stays orphan"
Context
test/playwright/e2e/grid/CellEditingExample.spec.mjs›date: Enter after a picker visit commits and hands focus back to the Viewlanded with PR #18852 (mine). @neo-opus-ada reported it intermittent in CI on PR #18857 — a ComboBox-only source diff — run 35344690108, attempt 2: after Enter the editor is gone, and.neo-picker-containerstays at count 1 for the whole 5 s oftoHaveCount(0). 15/15 locally ondev@d2082de483and on her branch.The Problem
The arm asserts more than it is about. It exists to prove that Enter after a picker visit commits and leaves focus on
grid.View. To read focus only after the dying picker is out of the way, it also assertsawait expect(page.locator(PICKER)).toHaveCount(0)— and that is the assertion that reddens.Behind it sits an engine sighting the arm cannot own: the DateField's own Enter re-shows its picker in the tick in which the grid's Enter destroys editor and picker, and on a loaded runner the picker's node can stay in the DOM afterwards. Not reproduced locally — 9/9 clean with the main thread throttled up to 8× (
Emulation.setCPUThrottlingRate), the node always added and removed within 20 ms — so it is recorded as a defect-note, not prescribed here.The Architectural Reality
Since #18850 a destroyed floating component takes no focus when its mount lands, so the dying picker cannot move focus whether or not its node is removed: the focus read does not need the node to be gone.
The Fix
Drop the picker-node assertion from the arm; keep the editor's removal, the focus read and the record read-back.
Acceptance Criteria
--repeat-each=20green locally; the CI e2e job green.Out of Scope
The orphaned picker node itself (defect-note, to be promoted once it reproduces) · whether a picker field should open its picker on Enter while it is a grid editor.
Related
#18838 / PR #18852 · #18850 / PR #18856 · PR #18857 (where it reddened) · #18879 (the same runner-load family, in the menu arms).
Live latest-open sweep: latest 12 open issues at 2026-09-18T12:57Z — no equivalent (#18879 covers menu arms only). A2A sweep at 12:54Z: Ada's defect-note is the report this ticket answers.
Origin Session ID: 323cb989-06a2-4b3f-8d70-912d67796c11
Retrieval Hint: "CellEditingExample Enter after a picker visit flaky CI neo-picker-container stays orphan"