Inconsistent behavior with `waitFor` when using fake timers globally vs per-test

[lingokids-carlos-garcia]

Describe the bug

waitFor and functions that use it behind-the-scenes like findBy can produce inconsistent results when fake timers are enabled globally versus per-test. This specifically affects tests that render components with async mount effects that trigger state updates. When fake timers are enabled globally in jest.config.js, the test behavior may differ compared to enabling them within individual test files using jest.useFakeTimers().

Example:

const MyComponent = () => {
  const [count, setCount] = React.useState(0);

  const myPromise = async () => {
    const response = await new Promise<number>((resolve) => resolve(1));
    setCount(response);
  };

  React.useEffect(() => {
    myPromise();
  }, []);

  return count > 0 && <Text testID="count">{count}</Text>;
};

Expected behavior

Tests produce the same result independently to enable fake timers globally.

Steps to Reproduce

A couple of branches have been created to demonstrate this case:

• Case: Test fails when enabling fake timers globally
• Case: Test passes when enabling fake timers on the test

Screenshots

Case: Test fails when enabling fake timers globally

Case: Test passes when enabling fake timers on the test

Versions

  npmPackages:
    react: 19.0.0 => 19.0.0
    react-native: 0.79.1 => 0.79.1
    react-test-renderer: 19.0.0 => 19.0.0

[lingokids-carlos-garcia]

I investigated if the issue might be related to the setImmediate function used by fake timers, but found that the code is actually using the real timers implementation.

react-native-testing-library/src/helpers/timers.ts

Lines 85 to 87 in 0f5cc2e

	const { clearTimeoutFn, setImmediateFn, setTimeoutFn } = runWithRealTimers(
	bindTimeFunctions,
	) as BindTimeFunctions;

Notably, this issue was discovered in a project that worked correctly with React 18 but started failing after upgrading to React 19.

[mdjastrzebski]

@lingokids-carlos-garcia which behavior do you consider correct , the global or per-test one? Why?

[lingokids-carlos-garcia]

Hey @mdjastrzebski 👋 ! Thanks for taking a look at this 🙇 !

@lingokids-carlos-garcia which behavior do you consider correct , the global or per-test one? Why?

Not sure if any of them should be considered incorrect, from my side, I'd say both are correct. Depending on your testing needs, you could define fake timers globally or per-test.

The problem I encountered involves the behavior of waitFor, particularly its internal flushMicroTasks function, which seems it has different behaviors based on when fake timers are enabled. Here's the key issue:

1. When fake timers are enabled globally:

   • The setImmediate function used in flushMicroTasks references the real timers version (i.e. this should be the original version).
   • This occurs because real timers are set during the import process (reference).

2. When fake timers are enabled per-test:

   • The setImmediate function still references the original version.
   • However, real timers don't need to be enabled in this case. This is because fake timers haven't been enabled at the time the testing library is imported.

If in both cases setImmediate points to the original version, I wonder why it produces different results.