Intersection Observer tests are returning incorrect results in Edge 86

[thejohnjansen]

I'm not sure exactly where is the best place to log this kind of an issue, but since it looks like it might be an infra issue, starting here.

I noticed that for the Intersection Observer test results, it appears that Edge is failing a significant number of tests. However, when I run them manually, Edge passes those tests. For example, the table says we pass 2/5 Display-None tests, but when I run the test file, I see we pass all of the tests.

I'm not exactly sure how best to debug something like this. I'm happy to do so, but I typically debug valid failures, not incorrect reporting.

[stephenmcgruer]

Going back to late 2019 I don't see any regression here: https://wpt.fyi/results/intersection-observer?diff&filter=ADC&run_id=410000005&run_id=623520006

So this likely always happened. Are you able to reproduce via wpt run? E.g. on Windows:

python wpt run --log-mach=- --channel=dev edgechromium [test name]

[stephenmcgruer]

See also https://web-platform-tests.org/running-tests/from-local-system.html#windows-setup for the basic Windows setup.

[foolip]

@thejohnjansen @mustjab do you know if the Edge setup we have is running headless or "headfull"? That's a difference that one might expect to lead to this.

[thejohnjansen]

@foolip great callout! I have been trying to repro and not seeing issues, so I will try headless.

[thejohnjansen]

@foolip @mustjab it looks like this might be linux vs windows, not headless vs headfull. When I run the tests manually, they all pass every time. When I run them automated via webdriver, they fail. Even if I try to refresh the page automatically, they still fail. If I refresh the page manually, they Pass. I'm going to play around with this now.

I know you didn't intend to, but thank you for reminding me that Chrome runs on Linux and Edge runs on Windows. :-)

[thejohnjansen]

I made a patch for this, but it's not pretty. I added a few setTimeout() functions to the intersection-observer/resources/intersection-observer-test-utils.js file in the function checkLastEntry().

Is there a recommended way to slow down a test like this without using a bunch of setTimeouts()?

Of course, these tests seem kind of hacky already given a comment in that file:

// Ideally, it should be sufficient to use requestAnimationFrame followed // by two step_timeouts, with the first step_timeout firing in between the // requestAnimationFrame handler and the task to deliver notifications. // However, the precise timing of requestAnimationFrame, the generation of // a new display frame (when IntersectionObserver notifications are // generated), and the delivery of these events varies between engines, making // this tricky to test in a non-flaky way.

[szager-chromium]

I'm pretty firmly against adding setTimeout's for this. We went through a pretty strenuous process to make sure these tests are deterministic and as exacting as possible about the timing of IntersectionObserver notifications. Adding a setTimeout fudge factor just papers over legitimate web compatibility issues, and it also makes the tests inherently slow (which is something we try hard to avoid). I think it's important to understand the root cause of the failures.