Fix relative time in pthreads [v2]

[niklasf]

This is #9799 + the requested trivial changes and some tweaks. Hope this helps to make progress towards landing the fix.

[welcome]

Thank you for submitting a pull request! If this is your first PR, make sure to add yourself to AUTHORS.

[juj]

Looking at the changes, agree that the relative clock skew computation was broken after the move from global state to Module-specific state. What strikes me curious is there is already a test for this skew in test_browser.py:

  # Test that emscripten_get_now() reports coherent wallclock times across all pthreads, instead of each pthread independently reporting wallclock times since the launch of that pthread.
  @requires_threads
  def test_pthread_clock_drift(self):
    self.btest(path_from_root('tests', 'pthread', 'test_pthread_clock_drift.cpp'), expected='1', args=['-O3', '-s', 'USE_PTHREADS=1', '-s', 'PROXY_TO_PTHREAD=1'])

I wonder why that test would pass even when the implementation regressed?

[niklasf]

Curious indeed. A major difference might be that the existing test uses -s PROXY_TO_PTHREAD. (I tried to verify that, but removing PROXY_TO_PTHREAD runs into issues with blocking on the main thread).

[kripken]

Thanks for the help here @niklasf! Looks perfect.

Yes, I think the issue with the old test was that it uses proxy-to-pthread. That avoided the bug somehow (I looked at this back then, but don't remember the details). I did verify that the new test was broken before the fix, and fixed with it.