Bugfix: Synchronous ping during render phase sometimes unwinds the stack, leading to crash (#25851)

I found this bug when working on a different task.

`pingSuspendedRoot` sometimes calls `prepareFreshStack` to interupt the
work-in-progress tree and force a restart from the root. The idea is
that if the current render is already in a state where it be blocked
from committing, and there's new data that could unblock it, we might as
well restart from the beginning.

The problem is that this is only safe to do if `pingSuspendedRoot` is
called from a non-React task, like an event handler or a microtask.
While this is usually the case, it's entirely possible for a thenable to
resolve (i.e. to call `pingSuspendedRoot`) synchronously while the
render phase is already executing. If that happens, and work loop
attempts to unwind the stack, it causes the render phase to crash.
This commit is contained in:
Andrew Clark 2022-12-21 23:55:49 -05:00 committed by GitHub
parent 2b1fb91a55
commit 5fcf1a4b4c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 86 additions and 2 deletions

View File

@ -3401,8 +3401,16 @@ function pingSuspendedRoot(
includesOnlyRetries(workInProgressRootRenderLanes) &&
now() - globalMostRecentFallbackTime < FALLBACK_THROTTLE_MS)
) {
// Restart from the root.
// Force a restart from the root by unwinding the stack. Unless this is
// being called from the render phase, because that would cause a crash.
if ((executionContext & RenderContext) === NoContext) {
prepareFreshStack(root, NoLanes);
} else {
// TODO: If this does happen during the render phase, we should throw
// the special internal exception that we use to interrupt the stack for
// selective hydration. That was temporarily reverted but we once we add
// it back we can use it here.
}
} else {
// Even though we can't restart right now, we might get an
// opportunity later. So we mark this render as having a ping.

View File

@ -4218,4 +4218,80 @@ describe('ReactSuspenseWithNoopRenderer', () => {
});
expect(Scheduler).toHaveYielded(['Unmount Child']);
});
// @gate enableLegacyCache
it(
'regression test: pinging synchronously within the render phase ' +
'does not unwind the stack',
async () => {
// This is a regression test that reproduces a very specific scenario that
// used to cause a crash.
const thenable = {
then(resolve) {
resolve('hi');
},
status: 'pending',
};
function ImmediatelyPings() {
if (thenable.status === 'pending') {
thenable.status = 'fulfilled';
throw thenable;
}
return <Text text="Hi" />;
}
function App({showMore}) {
return (
<div>
<Suspense fallback={<Text text="Loading..." />}>
{showMore ? (
<>
<AsyncText text="Async" />
</>
) : null}
</Suspense>
{showMore ? (
<Suspense>
<ImmediatelyPings />
</Suspense>
) : null}
</div>
);
}
// Initial render. This mounts a Suspense boundary, so that in the next
// update we can trigger a "suspend with delay" scenario.
const root = ReactNoop.createRoot();
await act(async () => {
root.render(<App showMore={false} />);
});
expect(Scheduler).toHaveYielded([]);
expect(root).toMatchRenderedOutput(<div />);
// Update. This will cause two separate trees to suspend. The first tree
// will be inside an already mounted Suspense boundary, so it will trigger
// a "suspend with delay". The second tree will be a new Suspense
// boundary, but the thenable that is thrown will immediately call its
// ping listener.
//
// Before the bug was fixed, this would lead to a `prepareFreshStack` call
// that unwinds the work-in-progress stack. When that code was written, it
// was expected that pings always happen from an asynchronous task (or
// microtask). But this test shows an example where that's not the case.
//
// The fix was to check if we're in the render phase before calling
// `prepareFreshStack`.
await act(async () => {
root.render(<App showMore={true} />);
});
expect(Scheduler).toHaveYielded(['Suspend! [Async]', 'Loading...', 'Hi']);
expect(root).toMatchRenderedOutput(
<div>
<span prop="Loading..." />
<span prop="Hi" />
</div>,
);
},
);
});