clock.performance.now returns incorrect value within clock.setInterval callback

• FakeTimers version : 7.1.2
• Environment : NodeJS
• Example URL : N/A; see test below
• Other libraries you are using: Jest

When passing fractional milliseconds to clock.setInterval and clock.tick, the return value of clock.performance.now is sometimes incorrect when called within the callback passed to clock.setInterval.

Here is an example test running within Jest:

import { Clock, install } from '@sinonjs/fake-timers';

let clock: Clock;

beforeAll(() => {
  clock = install();
});

afterAll(() => {
  clock.uninstall();
});

test('fake timers bug', () => {
  const timestamps: number[] = [];
  clock.setInterval(() => {
    timestamps.push(clock.performance.now());
  }, 16.6);

  clock.tick(16.6);
  clock.tick(16.6);

  expect(clock.performance.now()).toBe(33.2); // Passes
  expect(timestamps).toStrictEqual([16.6, 33.2]); // Fails
});

What did you expect to happen?

Each timestamp should be a multiple of 16.6.

What actually happens

The timestamp generated on the second iteration of the callback is off by 1 millisecond. The value 32.2 is returned rather than 33.2.

Interestingly, when calling clock.performance.now outside of the clock.setInterval callback, the correct timestamp is returned.

How to reproduce

See test sample above.

[rrogowski]

@fatso83 Thank you for the thoughtful reply! You raise a good point about writing code that does not conform to the spec. I will re-think the code I am writing, which will hopefully make the functionality easier to test. I really appreciate your time on this matter, and I’m happy to close this issue as a non-bug!

[fatso83]

Not totally sure what to say here. Code that uses a non-integer duration is essentially “wrong” (not spec-conforming), so this is a more general question of how would you test something that is not supported😃

I think your approach of testing the timers schedule updates at an approximately correct frequency should be fine, and it’s basically what you did above. In any case, most JS runtimes do not make hard realtime guarantees, so whether your timer in practice gets called 51 times per minute or 62 times per minute is totally browser/node/runtime specific. So you are only really testing that fake-timers schedules a call every parseInt(16.6) millisecond. Not sure if we can do better than that.