corelib/date_time_extremes_test is failing in JavaScript on Firefox,

[nshahan]

The tests

corelib/date_time_extremes_test RuntimeError (expected Pass)

are failing on configurations

dart2js-linux-firefox
ddc-linux-fireox

We can track our test failure with this issue but it seems like this is a bug in Firefox. I opened a report:
https://bugzilla.mozilla.org/show_bug.cgi?id=1944995

[nshahan]

This test is starting to flake on dart2js. Sometimes it passes and I have no idea why at this point. Still waiting for the fix to be released.

[lrhn]

Was looking at the test, and it may have some weak points.
Doesn't have to be why it fails here, but I can't say it isn't.
(This is calendar-stuff, nothing is guaranteed, or stable.)

The test contains lines like:

  Expect.throws(
    () => DateTime(dt.year, dt.month, dt.day, dt.hour, dt.minute, 0, 1),
  );

That assumes that the seconds and milliseconds are zero.
Because surely no time-zone differs by seconds!

Except some do. Fx. Danish time prior to April 1st 1893 is listed as -0:53:28, which applies to all dates since the beginning of time (aka 100,000,000 days before epoch).

void main() {
  // Prints: -271821-04-20 00:53:28.000
  print(DateTime.fromMillisecondsSinceEpoch(-8_640_000_000_000_000));
}

If the test assumes seconds, or even minutes in some cases, are zero for non-UTC time, then it's overly optimistic.

I think browsers, or the time zone data, have changed behavior "recently", because I haven't seen the -0:53:28 offset before. (But probably not that new either.)

In any case, I'll update the test to not have these "is going to be zero" assumptions. Maybe it makes a difference.

The same thing could be affecting in the Firefix issue. The TZ database contains:

Zone America/Los_Angeles -7:52:58 -	LMT	1883 Nov 18 20:00u

which applies since beginning of time, which could be the source of the 7 minutes 2 seconds. Somehow (it should still be an error to do new Date(-271821, 3, 19, 16, 0, 0, 0) with a time-zone offset of -7:52:58).

[a-siva]

I have approved this test failure for the dart2js-linux-firefox bot at least two times so far and I find that it somehow keeps getting undone and shows up as an error in the bot,

[nshahan]

As I understand our flake detection logic if a test changes from pass to fail it will get rerun some number of times on the same bot. If the rerun results are all the same then the status changes. If the rerun results differ then it gets marked as flaky. This makes me believe there is some interaction happening where this test will fail sometimes but it will fail consistently on that bot. So it just switches between Pass and Fail when this happens.

The Firefox update that includes the fix is scheduled to be released on April 1 so I'm inclined to just approve the failures and wait until then to see if we need further investigation.

[nshahan]

The test has been passing since we bumped the version of firefox to 137.