Context
@neo-opus-grace measured it while chasing the CI-only picker red behind #19265 (defect-note MESSAGE:d2406cc3, 11:45Z). With a +150 ms settle before the arm's second trigger click, the date arms failed 6/6 for her, even unthrottled. The plain spec failed 9/16 for her on dev at 4x CPU.
The Problem
test/playwright/e2e/grid/CellEditingPickerProjection.spec.mjs, EDITORS.date.open:
- Clicks the trigger.
- Picks
2024-12-15.
- Asserts the draft.
- Clicks the trigger again, meaning to open the picker again, as the docblock says: "A date draft is picked, and the picker opened again".
After a pick, though, the picker is still open: Neo.form.field.Date#hidePickerOnSelect defaults to false, and the fixture does not set it. The trigger toggles, so the second click closes the picker.
Measured at dev a87a89e293 (a probe logging the picker count):
| step |
picker count |
| after the pick |
1 |
| after 150 ms |
1 |
| right after the second click |
1 |
| 300 ms later |
0 |
The arm's "the picker is open over the draft" assertion therefore passes inside the close's latency. The scroll then starts from a picker that is already closing, so "no picker floats over the vanished field" holds whether or not projection works: the date arms witness nothing. Whether they also go red depends on timing. Here the original passed 16/16 at 4x CPU and 2/2 with the settle, while Grace saw 9/16 and 6/6; that difference is the timing dependence itself.
The Architectural Reality
- Every later step (scroll away, the picker goes with its field, reprojection keeps the draft) needs a picker that is genuinely open as its starting state.
- Product behavior is out of this ticket's reach: whether a grid date editor should close on pick is a separate call. The arm has to be deterministic under either answer.
The Fix
Reopen from a known-closed state:
- After the pick, assert the picker is open.
- Click the trigger and assert the picker is closed.
- Click the trigger again.
The picker the scroll starts from is then open and stays open.
Acceptance Criteria
Out of Scope
hidePickerOnSelect for grid date editors: a product question, with no ticket until someone decides it.
- #19265 / PR #19268 (the double picker node): a different mechanism, and it does not touch this spec.
Related
#19265 · PR #19268 · the defect-note above.
Live latest-open sweep: the latest 20 open neo issues at 11:5xZ, plus gh search for the spec's name across open issues and PRs. #19265 names the spec for its combo arm (a different mechanism), and PR #19268 does not touch the spec. A2A in-flight sweep: Grace's note hands the call to the spec's author; no competing claim. Memory Core sweep: no prior ruling on the reopen step. Own-assignment sweep: none covers it.
Origin Session ID: 1b945fcf-1142-475f-8007-ac18d51c069a
Context
@neo-opus-grace measured it while chasing the CI-only picker red behind #19265 (defect-note
MESSAGE:d2406cc3, 11:45Z). With a +150 ms settle before the arm's second trigger click, the date arms failed 6/6 for her, even unthrottled. The plain spec failed 9/16 for her on dev at 4x CPU.The Problem
test/playwright/e2e/grid/CellEditingPickerProjection.spec.mjs,EDITORS.date.open:2024-12-15.After a pick, though, the picker is still open:
Neo.form.field.Date#hidePickerOnSelectdefaults tofalse, and the fixture does not set it. The trigger toggles, so the second click closes the picker.Measured at dev
a87a89e293(a probe logging the picker count):The arm's "the picker is open over the draft" assertion therefore passes inside the close's latency. The scroll then starts from a picker that is already closing, so "no picker floats over the vanished field" holds whether or not projection works: the date arms witness nothing. Whether they also go red depends on timing. Here the original passed 16/16 at 4x CPU and 2/2 with the settle, while Grace saw 9/16 and 6/6; that difference is the timing dependence itself.
The Architectural Reality
The Fix
Reopen from a known-closed state:
The picker the scroll starts from is then open and stays open.
Acceptance Criteria
Out of Scope
hidePickerOnSelectfor grid date editors: a product question, with no ticket until someone decides it.Related
#19265 · PR #19268 · the defect-note above.
Live latest-open sweep: the latest 20 open neo issues at 11:5xZ, plus
gh searchfor the spec's name across open issues and PRs. #19265 names the spec for its combo arm (a different mechanism), and PR #19268 does not touch the spec. A2A in-flight sweep: Grace's note hands the call to the spec's author; no competing claim. Memory Core sweep: no prior ruling on the reopen step. Own-assignment sweep: none covers it.Origin Session ID: 1b945fcf-1142-475f-8007-ac18d51c069a