Patch 1.
The tests on Windows show that when the timer callback is queued behind the clock time it is queued and executed quite soon with 0-15ms delay which corresponds to default timer granularity. The timeout tolerance in the test is quite bigger to avoid a flaky test. Wine currently may execute the callback right away or delay it indefinitely depending on the time difference and current system time.
Patch 2. It turns out when the session is paused and then restarted and all the samples were delivered to sample grabber before restart, nothing is going to request the samples again when the session is restarted. I am attaching a test in a patch which hopefully confirms how that should work. I am not including that in the patches as the test involves delays (to make sure all the async events settle at certain points) and inherently flaky. Also, _ IMFStreamSink_GetEvent(stream, ...) in the test crashes on Win7 for some unknown reason.
[0001-mf-tests-Add-test-for-pausing-and-restarting-sample-.patch](/uploads/58c1738d48b3392e7ec430a0757210d0/0001-mf-tests-Add-test-for-pausing-and-restarting-sample-.patch)
-- v2: mf/samplegrabber: Request a sample when going from paused to running state. mf: Handle timer time behind clock time in present_clock_schedule_timer().