test(setup): observe promises in the ClipboardItem stub

[johnny4young]

Summary

Fixes the unhandled Canceled rejection that failed the Unit job on the main push after #201, while all 8,089 tests passed. The cause is the ClipboardItem stub in tests/setup.ts: it stores the promises it is given without observing them. It now observes them the way a real ClipboardItem does, and a regression test pins the behavior.

Root cause

1. The workaround. jsdom's user agent carries AppleWebKit without a Chrome or Safari token. When Monaco initialises its services it therefore installs its WebKit clipboard workaround: click and keydown listeners on the document body.
2. The promise. Each click passes a pending promise into new ClipboardItem({ 'text/plain': promise }), and the next click cancels the previous one.
3. The gap. A real ClipboardItem observes those promises. The stub only reads them in getType(), which nothing calls, so every cancellation became an unhandled rejection. Vitest fails the run on it even when every test passes.
4. Why it was intermittent. In the unit suite, Monaco gets initialised by JsonSyntaxOutput, which imports it without awaiting and gives up if the component has already unmounted. On a fast machine the tests finish first. On a slow runner the component is still mounted, Monaco initialises, and later clicks in the same file hit the workaround.

The CI run (35143874393) reported the rejection at monaco-editor/.../clipboardService.js:74, originating in tests/components/UtilityToolbarApply.test.tsx. The failed smoke artifact upload was only a consequence: the test step stopped before writing the smoke summary. Production is unaffected, because browsers use their real ClipboardItem.

Changes

• tests/setup.ts. The stub's constructor attaches a no-op rejection handler to each promise-like value. getType() still awaits the original promise, so a reader of a rejected type still sees the rejection; types, getType, supports and the writable global are unchanged.
• tests/clipboardItemStub.test.tsx (new).
  • Stub level. A promise rejected after the item is built leaves no unhandled rejection. A reader of a rejected type still gets the error. Resolved promises, strings and blobs stay readable, and unknown types still throw the MIME error.
  • Monaco level. Monaco is initialised the way JsonSyntaxOutput does it, and the test clicks three times with user-event. It asserts that the workaround wrote to the clipboard, so the guard fails loudly if Monaco ever drops it, and that no rejection went unhandled.

Type of change

• Bug fix
• New feature
• Refactor / internal
• Docs
• Build / CI / tooling

How was this tested?

• Regression test red before the fix. Against the original stub, the delayed-rejection case failed with Error: write cancelled and the Monaco case with two Canceled rejections. With the fix, all four cases pass.
• The failing CI path, made deterministic.
  • Temporary edit: UtilityToolbarApply.test.tsx initialises Monaco in a beforeAll, which is what a slow runner ends up doing.
  • With the original stub: its 10 tests passed alongside 6 unhandled Canceled rejections, the CI signature.
  • With the fixed stub: its 10 tests passed with none.
• Related suites. qrCode.test.ts, QrCodePanel.test.tsx, UtilityToolbarApply.test.tsx and DeveloperUtilitiesModal.test.tsx pass 77 tests.
• Unit. pnpm test -- --run passes 732 files and 8,095 tests with no unhandled errors: the counts on main plus the new file.
• Other gates. tsc --noEmit, typecheck:tests, lint including the telemetry audit and changelog:check pass. The new test and tests/setup.ts also type-check clean under a scratch config. knip reports only the existing website test dependencies.
• e2e and smokes. Not applicable: only the test environment changes.

Checklist

• pnpm test -- --run passes
• pnpm exec tsc --noEmit passes
• pnpm run lint passes
• pnpm run check:i18n and pnpm run check:i18n:copy pass (if UI/i18n touched) — no UI or i18n in this diff
• User-facing changes were verified in a running build (web or desktop) — no user-facing change
• Docs updated if shortcuts, execution, or workflow behavior changed

Risk and review notes

• No silencing. No global unhandled-error filter is added; only rejections the stub itself receives are observed, as a real ClipboardItem would.
• Test cost. The Monaco-level case loads Monaco in jsdom, which several other tests already do.
• Alternative not taken. Giving jsdom a Chrome-like user agent would stop Monaco installing the workaround, but it would change platform detection for every test.