test(renderer): keep the TaskProvider guard throw inside the component - #264

What happened

A full npm run test:coverage run failed once on a green branch with Error: useTaskContext must be used within an TaskProvider, pointing at taskManagerFlows.test.tsx:82. Line 82 is the test that deliberately asserts that guard throws, so the test itself was fine. The error was arriving as a run-level unhandled error, which is why it looked random and why it never showed up when the file ran alone.

Reproduction

I could not force a red run. 27 clean shuffled runs of the renderer-dom project (--sequence.shuffle with seeds 1001 through 1029, minus three runs I contaminated with my own probe file) all passed, and 17 full coverage runs all passed. So no failing seed to hand over.

What does reproduce, on every single run, is the error itself. Every green full coverage run printed this at the top of its output, twice, with no test file attributed to it:

Error: useTaskContext must be used within an TaskProvider
    at useTaskContext (src/renderer/src/contexts/TaskManagerContext.tsx:404:11)
    at tests/renderer-dom/taskManagerFlows.test.tsx:82:35
    ...
    at HTMLUnknownElement.callCallback (node_modules/react-dom/cjs/react-dom.development.js:4164:14)
    at innerInvokeEventListeners (node_modules/jsdom/lib/jsdom/living/events/EventTarget-impl.js:360:16)

That was the only stderr output the whole suite produced. Same message, same stack, same file, same line as the failure. It just usually lands as noise instead of as a failure.

The mechanism

The test rendered a component that throws and let the throw escape the render:

expect(() => renderHook(() => useTaskContext())).toThrow(/must be used within an TaskProvider/)

React's development build replays a failed beginWork through invokeGuardedCallbackDev, which dispatches a synthetic event on a detached node so the throw reaches the host's uncaught error machinery and devtools can pause on it. That is the HTMLUnknownElement.callCallback frame in the stack. jsdom then catches the listener exception and fires an error event on window, and since nothing cancels it, jsdom reports it to the console. React attempts the render twice, hence two copies.

Vitest's jsdom environment installs its own window error listener. When an uncancelled error event arrives and no user listener is currently registered, it calls preventDefault() and re-emits the error as process.emit("uncaughtException", ...). I confirmed that route with a throwaway test that dispatches one uncancelled error event and nothing else:

⎯⎯⎯⎯⎯ Uncaught Exception ⎯⎯⎯⎯⎯
Error: scratch escalation probe
This error originated in "tests/renderer-dom/zzScratchProbe.test.tsx" test file. It doesn't mean
the error was thrown inside the file itself, but while it was running.
 Test Files  1 passed (1)
     Errors  1 error

The run exits non-zero, the file passes, and the error is pinned on whichever file was running. That is exactly the shape of the reported failure.

The only thing standing between the guard test and that outcome is that React registers its own transient window error listener around the dispatch, which bumps Vitest's user-listener count to one for the duration. I instrumented the window to watch it, and measured the event arriving at depth 1 with defaultPrevented false every time. That count is a single window-wide integer, and Vitest decrements it on any removeEventListener("error", ...) whether or not a matching add ever happened. I watched it go to -1 during environment teardown. So the suppression is incidental, not designed, and it holds only as long as nothing else in the same window touches error listeners at the wrong moment.

The fix

Catch the throw inside the component so React never handles it. No guarded replay, no synthetic event, no window error, nothing for Vitest to escalate. The guard on TaskManagerContext.tsx:404 is still executed and its message still asserted, so coverage is unchanged.

configContextSlices.test.tsx has the same kind of test and is already safe. It goes the other way and silences the escape with a console.error mock plus a listener that cancels the event. That works, so I left it alone.

The pin

The test now spies on console.error and asserts it was never called. React logs "The above error occurred in ..." for every render throw it has to handle itself, so anyone who reverts to letting the throw escape gets a failure in this file rather than a mystery unhandled error somewhere else in the run. Verified by putting the old escaping version back with the spy in place:

× throws when used outside a TaskProvider
AssertionError: expected "error" to not be called at all, but actually been called 1 times
 Test Files  1 failed (1)

Gates

npm run typecheck, npm run lint:ci (0 errors, the same 15 pre-existing warnings) and npm run format:check all pass. Three consecutive npm run test:coverage runs came back green at 137 files and 1633 passed with 2 skipped, coverage at 92.58 lines, 89.8 statements, 92.03 functions and 94.05 branches, all comfortably over the floors. The run output is now clean, where before every run printed those two error blocks.