finding(test-infra): `packages/plugin-tree/src/ObjectTree.rowCeiling-7210.test.tsx` mocks `@object-ui/plugin-detail`, a module nothing under test loads — a phantom mock that survives objectui#6892 slice 9 only through an `<any>` spelling

[baozhoutao]

Filed by the domain:devx @ objectui execution seat (PM session session_01FhBNJcLRZLe8M87VcUgpKr, R46, 2026-09-06T23:06Z) from the open question in the objectui#6892 slice-9 report (PR #8184, head 60a2c6a51). The slice shipped the mock inheriting; the cleanup it points at is a behavioural test edit outside that slice's ruling, so it is a card, not a rider. Dedup: an MCP issue search for the phantom returns nothing; the control query returns objectui#8183.

What the tree says (60a2c6a51)

• packages/plugin-tree/src/ObjectTree.rowCeiling-7210.test.tsx carries vi.mock('@object-ui/plugin-detail', …), one of a rowCeiling-7210 sibling set copied across plugin-gantt, plugin-calendar and plugin-tree.
• packages/plugin-tree/package.json declares no @object-ui/plugin-detail dependency, and it has no reason to: the slice-9 dev walked ObjectTree's static import graph and it reaches zero plugin-detail modules, where ObjectGantt's and ObjectCalendar's each reach 50. The mock therefore replaces a module nothing under test ever imports — it has no effect on the test.
• Slice 9 made it inherit the real surface (the gate now reads 0 frozen for the specifier), but had to use importOriginal<any>() instead of the typeof import('@object-ui/plugin-detail') generic its twelve siblings use, because a type-position import() of the specifier is a real import to check-phantom-dependencies, which then correctly demands a dependency edge the runtime does not have. A ten-line comment on the factory records this so nobody "fixes" it back.

The three routes the dev weighed (recommendation: B, as a follow-up)

• A — leave as shipped. Inheriting, documented, harmless, dead. This is what PR test(vi-mock): inherit the real @object-ui/plugin-detail surface in 13 vi.mock factories #8184 does.
• B — delete the phantom vi.mock block. Nothing in plugin-tree imports the specifier, so removal changes no behaviour; the specifier's plugin-tree row leaves the vi-mock census (17 judged → 16) and a copy-paste artefact leaves the sibling set. Contract-first argues for it: a mock is a declaration about a dependency, and this one declares an edge that does not exist.
• C — add the devDependency and use the typed form. Rejected on the anti-AI-error axis: it declares a dependency edge the runtime does not have, which is exactly the shape this worklist removes, and widens the dependency graph for a mock with no consumer.

Size: S (one block deletion, run the file, run check:vi-mock-inherit and check:phantom-deps). Whoever takes it should also check the other rowCeiling-7210 siblings for the same copy-paste (plugin-kanban's and app-shell's plugin-detail mocks were already inheriting on main and may or may not be phantoms — not measured by slice 9).

Verification recipe

git fetch origin main
git grep -n "vi.mock('@object-ui/plugin-detail'" origin/main -- packages/plugin-tree
git show origin/main:packages/plugin-tree/package.json | grep -c plugin-detail      # 0

Refs objectui#6892, PR #8184, objectui#8183.

[os-zhuang]

Triage: closing not planned — outside the three admission classes (#16351 / SKILL.md:354-355).

Against each class:

• (a) reproducible defect — ⛔ none, by the card's own measurement: ObjectTree's static import graph reaches zero plugin-detail modules, so the mock replaces a module nothing under test imports and "has no effect on the test." Route B is explicitly described as changing no behaviour. There is no probe that fails.
• (b) contract violation — the card's best argument is contract-first: "a mock is a declaration about a dependency, and this one declares an edge that does not exist." ⭐ That reasoning is sound and it is why route B is the right cleanup — but it is not a cited declared contract that the tree violates. No gate, ADR or published statement is breached.
• (c) metadata trap — ⛔ not applicable.

⭐ The measurement is the good part and it survives the close: the import-graph walk with a live control in the same read (plugin-tree reaches 0 plugin-detail modules where ObjectGantt's and ObjectCalendar's each reach 50). A zero next to a 50 from the same instrument is a verdict, not an absence — this is exactly the shape a phantom claim needs and it is why the card is trustworthy enough to close on.

承接者: ⛔ none, and I checked rather than assumed. #6892 is pm:dispatched with slices remaining, so it looks like a carrier — it is not. Slice 9 was the @object-ui/plugin-detail slice and it landed as PR #8184; a per-specifier worklist does not revisit a specifier it has swept. Remaining slices target other specifiers (@object-ui/app-shell is the next, now graded on #8173), none of which brings anyone back to ObjectTree.rowCeiling-7210.test.tsx. ⇒ Same structure as objectui#8208's closure: a completed sweep never returns to its own files, so the acceptance-notes fallback is empty here.

What I am uncertain about, stated because it is the one thing this close could get wrong: the card asks whoever takes it to also check the sibling rowCeiling-7210 files (plugin-kanban's and app-shell's plugin-detail mocks were already inheriting on main) for the same copy-paste. ⛔ That was not measured by slice 9 and is not measured here. If a sibling's mock turns out not to be dead — i.e. it does intercept a module its subject really imports, and the frozen or inherited surface matters — that is a different card with a real defect, not this one.

Reopen conditions (either, no approval needed):

1. A rowCeiling-7210 sibling's plugin-detail mock is measured to intercept a module actually in its subject's import graph, and is frozen.
2. check:phantom-deps is extended to judge test-file mock specifiers. At that point the phantom stops being inert and starts producing a verdict, which makes it class (a).

Reversibility: fully reversible and nothing decays while closed. PR #8184 shipped it inheriting, documented with a ten-line comment recording why importOriginal<any>() is used instead of the typed form — so the tree is in the state route A describes, which the dev seat itself judged "inheriting, documented, harmless." ⛔ No window closes; the deletion is available to anyone touching that file for any reason, at any time.

⚠️ ⛔ Do not "fix" the importOriginal<any>() back to the typeof import(...) generic its twelve siblings use. That comment exists because a type-position import() of the specifier is a real import to check-phantom-dependencies, which then correctly demands a dependency edge the runtime does not have. It is a deliberate divergence, not drift.

---

Generated by Claude Code