Cannot read property 'return' of null

[jquense]

Running into a strange issue that i cannot reproduce, it happens when running tests in Karma using 16.3.0-alpha.3

The issue is here:

react/packages/react-reconciler/src/ReactFiberScheduler.js

Lines 864 to 877 in 208b490

	if (nextUnitOfWork === null) {
	// This is a fatal error.
	didFatal = true;
	onUncaughtError(thrownValue);
	break;
	}
	if (__DEV__ && replayFailedUnitOfWorkWithInvokeGuardedCallback) {
	const failedUnitOfWork = nextUnitOfWork;
	replayUnitOfWork(failedUnitOfWork, isAsync);
	}
	const sourceFiber: Fiber = nextUnitOfWork;
	let returnFiber = sourceFiber.return;

My test is throwing an error because of a missing onChange handler when rendering a checkbox with checked but no handler. What appears to be happening is that nextUnitOfWork is not null and misses the first branch, replayUnitOfWork is run, after which nextUnitOfWork is null and sourceFiber doesn't exist and so access of return fails.

I've tried for a while to reproduce it, the following codesandbox is a simplified version of the code that is throwing, but the error isn't present there :/ https://codesandbox.io/s/0332y243zp

[acdlite]

Hmm, I was concerned this might happen. We assume that replayUnitOfWork will be idempotent but looks like you've found a case where it's not. I'll take a look at this.

[acdlite]

Changing line 877 to

const sourceFiber: Fiber = failedUnitOfWork;

isn't a sufficient fix because if it's not equal to nextUnitOfWork, that means the stack is misaligned.

One option is to add an invariant that throws if nextUnitOfWork !== failedUnitOfWork. Usually we avoid DEV-only invariants, but since this should only be possible in an error case, anyway, maybe we can get away with it.

[gaearon]

Repro case from #12497: https://jsfiddle.net/thysultan/Luktwrdm/264/

[gaearon]

Btw, there's another potentially related bug that happens if you remove this feature flag override that manifests as an Unexpected pop warning. Maybe we should change this test suite to use the feature flag that people use in real bundles? This doesn't leave me very confident.

[gaearon]

Added failing tests in #12508.

[gaearon]

Fixed in #12508

[onpaws]

Cheers, this PR just saved us from what would have been a confusing day!

[JeffUberman]

I've upgraded to React v16.3.2 and am seeing a similar error:

It's referencing a similar area of the React codebase, without a null check: