[wptrunner] Cycle testdriver event loop when testharness isn't loaded

[jonathan-j-lee]

Workaround for https://crbug.com/340662810 that seems to work without an increase in test runtime or regressions (https://crrev.com/c/5698330/9). On Windows (the affected platform):

• With patch: 225 min total shard runtime, 1873 regressions, 322 unexpected timeouts
• Without patch: 225 min total shard runtime, 1888 regressions, 326 unexpected timeouts

Waiting for the load event doesn't seem to work (https://crrev.com/c/5698330/7).

[WeizhongX]

Thanks for the fix.

This does not happen to a lot of tests, so should not have an impact on total test time.

[jonathan-j-lee]

This does not happen to a lot of tests

I was running into a lot of failed flakiness checks, so hopefully this should mitigate that.

should not have an impact on total test time.

My worry was that testharnessreport.js would never load correctly, which would cause the executor to spin indefinitely. But as shown, that does not seem to be the case.

[WeizhongX]

This does not happen to a lot of tests

I was running into a lot of failed flakiness checks, so hopefully this should mitigate that.

should not have an impact on total test time.

My worry was that testharnessreport.js would never load correctly, which would cause the executor to spin indefinitely. But as shown, that does not seem to be the case.

If this happens, will the test timeout?

[jonathan-j-lee]

If this happens, will the test timeout?

Hypothetically, it would, I think, since even non-testdriver testharness tests use the message queue/testdriver loop to signal completion. The executor process would report EXTERNAL-TIMEOUT and need to be killed by its TestRunnerManager.