This class exposes Wait()-Set() logic to synchronize events.
- There is a bug in checking EventWrapper::Wait() as it returns [1,2]. Negating
these values cause us to always pass timeout checks.
- There is a general problem in this class with waiter. There are 2 scenarios:
1) Lock()-Unlock()-DisplaysReconfigured()
In this scenario, Wait() in DisplaysReconfigured() immediately passes as event
is already signaled. Next Lock() call won't continue until Set() is called in
DisplaysReconfigured(). This blocks capture thread from accessing display until
reconfiguration completes.
2) Lock()-DisplaysReconfigured()-Unlock()
In this scenario, Wait() in DisplaysReconfigured() passes when Unlock() called.
Capture thread accesses display while reconfiguration happens. Note that we are
only delaying the OS delegate thread here. As an experiment, adding Sleep() in
DisplaysReconfigured() results in no change, because it looks like OS uses this
thread for only delegates but not for the actual display switch.
Overall, (1) doesnt seem necessary as (2) already accesses display while
reconfiguration happens. (2) doesn't seem necessary as blocking system delegate
thread doesn't help. Therefore, I changed the class to only protect from race
condition on |desktop_configuration_|.
Bug: chromium:796889
Change-Id: I37263305e5ac629e21ff9e977952cf4a21bae19f
Reviewed-on: https://webrtc-review.googlesource.com/c/108560
Commit-Queue: Emircan Uysaler <emircan@webrtc.org>
Reviewed-by: Tommi <tommi@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#25437}
eaf337a141