Run the front-end unit tests from the repo root, with one storage for every Node

[Babissimo]

ClickUp: 2.4 Run the front-end unit tests from the repo root, with one localStorage stub. Also closes the Node 25+ half of 123zgec36re.

Stacked on #613, at the top of the stack main ← #595 ← #602 ← #620 ← #607 ← #617 ← #613 ← #596, which lands in one deploy.

Why

• There was no root Vitest config or root test script. A bare vitest from the root collected e2e's Playwright specs and ran the dashboard tests without dashboard/vite.config.js (no jsdom, globals, jest-dom or @edsc/timeline alias): about eleven false failures. The only documented way to run everything was a chained one-liner in ONBOARDING.
• Node 25 and later define localStorage and sessionStorage globals that hide jsdom's (localStorage is undefined without --localstorage-file). Eight test files carried their own copy of an in-memory Storage to get round it, and LiveAircraftMap.test.tsx, which used the ambient one, failed on any local Node 25+ while CI's Node 20 passed.

What

Two commits:

1. One storage for every dashboard test. dashboard/src/test/setupStorage.ts, a second setupFiles entry, installs jsdom's own localStorage and sessionStorage on window before collection and again before every test, emptied each time. So every Node sees the same implementation, and a test that swaps in a throwing storage cannot leak it into the next. The eight copies go; tests seed and read window.localStorage directly. The throwing-storage stubs stay, as they test a different thing.
2. Root commands. A root vitest.config.js runs every workspace whose test script runs Vitest (dashboard, packages/shared) as a project under its own config. The root package.json gains test, typecheck, lint, build and check, the last running the four in web-build's order. This is what 2.2's just check is to call. ONBOARDING's one-liner is replaced.

CI is unchanged: web-build still runs each workspace's own scripts from its matrix.

Verification

• Root npm test on Node 26.9 (the laptop), with no NODE_OPTIONS: 119 files, 1388 tests, all passed.
• Node 20 is CI's, and web-build runs these same suites on it, so this PR's run is the Node 20 result. A local run in node:20-alpine at a load average above 250 failed only on timeouts (13 at the 5 s limit, one findByRole), all in tests that pass on Node 26 and none touching storage, plus scalarTheme.test.ts, which reads backend/ and so could not import from the copy that run used.
• Control: with the setup file emptied, LiveAircraftMap.test.tsx and usePersistedState.test.tsx fail 11 tests on Node 26 with Cannot read properties of undefined (reading 'setItem').
• The changed files also pass under --pool=vmThreads.
• vitest list from the root collects 116 dashboard and 3 shared files, and no e2e specs.
• npm run typecheck and npm run lint pass (one pre-existing set-state-in-effect warning).

🤖 Generated with Claude Code

[Babissimo]

Both review findings are addressed in 2251bb2.

1. Silent fallback in setupStorage.ts. The setup file now throws when the environment is jsdom (recognised by jsdom's default user agent, … jsdom/<version>) but globalThis.jsdom is missing. A file that opts into another environment still loads, since it has no jsdom user agent. Checked both ways: a copy reading a missing global failed theme.test.tsx with the new error, and a @vitest-environment node probe file passed.

2. Duplicated matchMedia stub. All ten copies now import one helper, dashboard/src/test/matchMedia.ts (its own commit). It is a helper the tests call, not a default in the shared setup file: ThemeContext treats a missing matchMedia as a real case (jsdom, any non-browser render), and a test that doesn't stub it exercises that path. A setup-file default would remove that path from every test at once. The helper installs through vi.stubGlobal, so LiveAircraftMap's unstubAllGlobals teardown still removes it.

[claude]

Claude finished @jehanazad's task in 2m 10s —— View job

---

Tasks

• Gather context (PR diff, changed files, prior review comments)
• Understand the request (run /code-review:code-review on this PR)
• Run code review
• Post findings

Review

Isolated this PR's own diff (aa67e90^..128adab, the three commits unique to this branch — the rest of the branch is the stack it's built on) and independently re-checked it against the 18 files/+150/-285 the PR describes.

• vitest.config.js: manifest(dir).scripts?.test ?? "" filtered against /\bvitest\b/ correctly keeps dashboard and packages/shared (both "test": "vitest run") and drops e2e (no test script, only test:e2e*). Confirmed vitest: ^4.1.4 is the version installed at the root — projects is the correct Vitest 3+ config key for this (the renamed replacement for the old workspace option), so there's no version mismatch with the config shape used.
• setupStorage.ts: the jsdom-UA guard only throws when globalThis.jsdom is missing and the environment's UA matches jsdom's own (/\bjsdom\//), so a file that opts into @vitest-environment node is unaffected, as the PR's follow-up comment claims. The beforeEach re-install()s before clearing, which is required because the "throws on every access" stubs in theme.test.tsx, sidebarCollapse.test.tsx, etc. all replace window.localStorage with configurable: true, letting setupStorage's Object.defineProperty reclaim it on the next test — verified this is consistent across every such stub left in the diff, no non-configurable replacement that would break the recovery.
• CI stays decoupled from the new root config: web-build's matrix (.github/workflows/ci.yml lines ~350–365) runs npm test -w <workspace>, which executes that workspace's own test script from inside its own directory — it never touches the new root vitest.config.js, so the "CI is unchanged" claim holds.
• Mechanical test refactors (appearanceSwitch, errorBoundary, frontDoor, mapShell, mapThemeBinding, physicsRoute, signOutLanding, signedOutChrome, sidebarCollapse, theme, usePersistedState, LiveAircraftMap): read every file's diff in full. All are 1:1 replacements of local storage/matchMedia stubs with window.localStorage / the shared stubMatchMedia helper — no change in seeded values, assertions, or cleanup order. Grepped the whole dashboard/src tree for leftover stubStorage definitions or inline matchMedia mocks: none remain.
• Root package.json: new scripts block adds no new dependencies, so no package-lock.json update is needed; check runs lint && typecheck && test && build, matching web-build's per-workspace order as the PR states.

This matches the outcome of the two earlier review passes already in this thread (the 2251bb29 fixes for the silent-fallback and duplicated-matchMedia findings, and the subsequent independent pass) — no new issues found on this pass either. As in that prior pass, I did not execute the suite here (no dependency install in this environment); this is a static review, and the PR's own verification section (Node 26 and Node 20 runs, the empty-setup-file control, --pool=vmThreads) covers the empirical side.

No issues found.
· branch web/root-unit-tests