Consider following situation. A class owns a timer. In destructor of that class we call .cancel() asynchronous on timer before it's destruction.
Now timer is executing cancel in it's own internal thread, while it's doing that destructor of timer is called from owner's destructor. Timer destructor enqueues stop notification. If that enqueue is happening just after while loop from cancel notification, stop notification is gonna be dropped and timer will never stop.
Fix: Add new method in TimedNotificationQueue which will return a notification regardless of the time it needs to be executed.
Get number of pending tasks in the queue. Flush out that many notifications from queue while taking special consideration of pending Stop and Cancel notifications.
Add test for new method in TimedNotificationQueue and fix cancel all tests to actually check if notification got executed.
fixes#3986
Timer is implemented with internal queue. If a user wants to cancel all pending tasks it can call .cancel to schedule CancelNotification.
As a part of processing of CancelNotification it will just flush the whole queue. It does have special processing for StopNotification so that Timer destruction doesn't get blocked.
Now if we first schedule async cancel and before this first cancel is processed we schedule another cancel but this time a sync second one will block because it is never notified that all tasks are canceled, _finished event is never set on that flushed CancelNotification.
Fix: add diffrent processing in case of CancelNotification to set all of it's _finished events.
Also add a test for this situation.
