fix(scripts): the vite resolve oracle stops writing its scratch root into the swept repo root (objectui#9468)

[os-try-charles]

Fixes #9468

The race, reproduced before it was repaired

viteResolves() in scripts/__tests__/vite-objectstack-spec-dist.test.ts made its throwaway Vite build root with mkdtempSync directly in the repo root and removed it again seconds later. textFootprint() in scripts/check-i18n-dead-keys.mjs sweeps that same repo root with a single grep -rFn pass. In the same shard, grep enumerates the scratch directory, fails to descend into it, and exits 2 — and textFootprint()'s catch rethrows every status but 1 deliberately, its comment saying why. The shard dies with ~10,500 passing tests in it.

Driving the real textFootprint() over the real repo root, three legs:

leg	scratch dir	result
control	none present	rounds 3, ok 3, status2 0
failing	.vite-oracle-9408-* in the repo root	rounds 3, ok 0, status2 3
after the fix	.vite-oracle-9408-* under node_modules/	rounds 3, ok 3, status2 0
failing	.tmp-citation-gate-* in the repo root	rounds 3, ok 0, status2 3
after the fix	.tmp-citation-gate-* under node_modules/	rounds 3, ok 3, status2 0

The failing leg reproduces the CI message character for character but for the random suffix:

grep: /home/user/objectui-issue-9468/.vite-oracle-9408-tzczpE: No such file or directory
Serialized Error: { status: 2, … }

Why the scratch root moved to node_modules/ rather than to os.tmpdir()

The card offered two shapes. Both were measured; neither is the one that landed, and the reason is a constraint the producer already documented.

os.tmpdir() breaks the oracle. The docstring on viteResolves() states that the root lives inside the repository because bare-specifier resolution walks up from the importer looking for node_modules, and from /tmp there is none. Measured, not inherited — moving it to os.tmpdir() and running the suite:

AssertionError: Vite oracle resolved nothing — broken instrument: expected null to be truthy
 Test Files  1 failed (1)   Tests  1 failed | 39 passed (40)

The suite's own anti-vacuity control catches it, which is the good case; the bad case is that this is the exact failure direction that reads as a clean pass.

Adding .vite-oracle-* to the sweep's skip set fixes one scanner and leaves the class alive. That set is a whitelist, and objectui#9201 already answered this same class that way once — this card is its recurrence.

repoRoot/node_modules is the placement that satisfies both constraints at once: still inside the repository, so the walk-up finds a node_modules; and inside a directory every repo-wide scanner under scripts/ already skips, so nothing transient placed there is walked by any of them. scripts/__tests__/check-action-ref-convention.test.ts already puts its throwaway root in the same place.

The suite stays green at the new placement: Test Files 1 passed (1) · Tests 40 passed (40).

Two pins, each shown to fail

Both ablations mutated the tree on disk, proved the mutation landed before reading any result, and restored from HEAD with git diff HEAD empty afterwards.

pin	ablation	result
the oracle's scratch root is never a repo-root entry	put the placement back in the repo root	red: expected '.vite-oracle-9408-fq9i4U' to be 'node_modules'
the sweep never descends into node_modules	drop node_modules from TEXT_SWEEP_SKIP_DIRS	red: expected [ …(2) ] to deeply equal [ 'packages/x/src/Reader.tsx' ]

The second ablation earned its keep. The first draft of that pin spelled its fixture node_modules/some-dep/dist/index.js, and dist is in the skip set too — so the ablation left the pin green, pinning nothing. The fixture path now carries exactly one excluded segment, and the test comment records why.

Verification

check	verdict
vitest run --project unit on the two deciding suites	Test Files 2 passed (2) / Tests 175 passed (175)
vitest run --project unit on the citation gate's suite	Test Files 1 passed (1) / Tests 34 passed (34)
vitest run --project unit scripts/__tests__/	Test Files 169 passed | 2 skipped (171) · Tests 4870 passed | 2 skipped (4872)
pnpm type-check:scripts	exit 0; tsc --listFiles confirms all three changed files are in the program
pnpm check:control-bytes	OK (scanned 7785 tracked text file(s))
pnpm check:test-path-roots	OK
pnpm check:new-line-citations	0 new citation(s) -> exit 0
pnpm check:i18n-dead-keys	exit 0
node scripts/check-changeset-presence.mjs	No source or published contract of a released package changed in this range, so no changeset is owed.

ESLint over the three changed files: exit 0, 0 errors. The citation gate's file reports 0 warnings; the oracle's file carries 5 no-explicit-any warnings, all at lines outside every hunk this branch adds (warnings at 117/935/960/1112/1146, hunks at 142-177/188-189/193/220-235), so all pre-existing. This is a narrowed run, not the repo-wide one: the repo-wide lint is CI's, and the config carries no type-aware linting, so a change inside these three files cannot move the verdict on a file the branch does not touch.

All figures above were measured at 73dfdda84, the branch head this PR carries.

Acceptance notes

This branch repairs both repo-root scratch directories, so the claim is "removes the class as it exists in this tree" rather than "removes one instance".

scripts/__tests__/check-new-cross-file-line-citations.test.ts built the throwaway git repo for its exits 0 on findings under report-only and 1 under --strict test the same way, with mkdtempSync straight into the repo root. It is not a theoretical twin — the same reproduction against that prefix fails identically:

rounds 3, ok 0, status2 3
grep: …/.tmp-citation-gate-pF6w0r: No such file or directory

The same one-line move repairs it, and the suite is unchanged at the new placement: Test Files 1 passed (1) / Tests 34 passed (34) — including the git init that scratch repo performs, which creates its own repository wherever it is put.

⚠ That producer gets no positional pin of its own. The thing that would pin every producer at once is a gate refusing repo-root scratch state, and that is a new verification surface — a decision, not this card's repair. It stays recorded on objectui#9468 for its own card.

Noted, not filed — neither .vite-oracle-* nor .tmp-citation-gate-* is in .gitignore (only .objectui-tmp is). That is now moot for these two, since node_modules is ignored, but it stays recorded as the same decision the gate above would be: what the tree owes is a rule about where scratch state may live, not two more ignore entries. Carrier: the class card.

---

Generated by Claude Code

---

Generated by Claude Code