Surfaced while implementing #6768 (PR #6847). That card converted the 25 frozen sites; this is the property of the tree that produced them, which the fix deliberately did not touch.
The gap
A vi.mock factory that hand-lists the exports it returns freezes the mock export surface. The next export any widely-imported module reads at module scope then kills that file at COLLECTION: zero failed assertions, the tests never run, and the suite reads like flake. PR #6847 demonstrates the failure and its removal with a two-leg ablation.
Nothing prevents the 26th one. A new test file written with a hand-listed factory reintroduces the defect the same day, and the class reopens silently — the cost is paid later, by whoever adds the next export, in a red suite that does not point at them.
Why a grep will not do it (measured on #6768)
The card was written from a grep for the literal importOriginal and counted 36 sites. The true frozen count is 25. Eleven files already inherit the real surface under a different spelling and need nothing:
- 9 in
plugin-dashboard spread await vi.importActual('@object-ui/react') from a zero-parameter factory.
EnvironmentListToolbar.test.tsx names the callback param importActual.PageView.test.tsx names it orig.
So the recogniser has to be semantic — does the factory OBTAIN the real module (its callback param under any name, or vi.importActual of the same specifier) and SPREAD it — not a substring match. A name-based gate would demand edits to 11 already-correct files, and would be deleted by the first person it annoyed.
Shape of a fix (not a decision, for triage)
A gate in the family of the existing scripts/check-vi-mock-specifiers.mjs (same defect direction: a mock that is silently not doing what it looks like it does). It is green at rest today — PR #6847 leaves 0 frozen sites out of 106 — so it would start as a ratchet rather than a cleanup.
Open question for triage: whether to scope it to widely-imported workspace specifiers or to every vi.mock factory. The second is stricter but there are legitimate whole-module replacements (vi.mock('./ObjectCalendar', ...) in plugin-calendar/src/registration.test.tsx is a deliberate one, and its own comment explains why).
Generated by Claude Code
Generated by Claude Code
Surfaced while implementing #6768 (PR #6847). That card converted the 25 frozen sites; this is the property of the tree that produced them, which the fix deliberately did not touch.
The gap
A
vi.mockfactory that hand-lists the exports it returns freezes the mock export surface. The next export any widely-imported module reads at module scope then kills that file at COLLECTION: zero failed assertions, the tests never run, and the suite reads like flake. PR #6847 demonstrates the failure and its removal with a two-leg ablation.Nothing prevents the 26th one. A new test file written with a hand-listed factory reintroduces the defect the same day, and the class reopens silently — the cost is paid later, by whoever adds the next export, in a red suite that does not point at them.
Why a grep will not do it (measured on #6768)
The card was written from a grep for the literal
importOriginaland counted 36 sites. The true frozen count is 25. Eleven files already inherit the real surface under a different spelling and need nothing:plugin-dashboardspreadawait vi.importActual('@object-ui/react')from a zero-parameter factory.EnvironmentListToolbar.test.tsxnames the callback paramimportActual.PageView.test.tsxnames itorig.So the recogniser has to be semantic — does the factory OBTAIN the real module (its callback param under any name, or
vi.importActualof the same specifier) and SPREAD it — not a substring match. A name-based gate would demand edits to 11 already-correct files, and would be deleted by the first person it annoyed.Shape of a fix (not a decision, for triage)
A gate in the family of the existing
scripts/check-vi-mock-specifiers.mjs(same defect direction: a mock that is silently not doing what it looks like it does). It is green at rest today — PR #6847 leaves 0 frozen sites out of 106 — so it would start as a ratchet rather than a cleanup.Open question for triage: whether to scope it to widely-imported workspace specifiers or to every
vi.mockfactory. The second is stricter but there are legitimate whole-module replacements (vi.mock('./ObjectCalendar', ...)inplugin-calendar/src/registration.test.tsxis a deliberate one, and its own comment explains why).Generated by Claude Code
Generated by Claude Code