Re: [PATCH v5 0/2] MR7833: mf: Update state and start clock for both paused and stopped.

[Nikolay Sivov (＠nsivov)]

On Thu Apr 24 20:50:29 2025 +0000, Brendan McGrath wrote:
> I think Windows' approach has merit if we are requesting a new start
> time, but for resumption I don't think it does. They don't perform any
> of the flushing in this scenario. So if the source changes the position
> significantly, then all the sinks and MFTs will have stale data. I guess
> they just rely on the source not doing that on resumption.
> But the other issue is, as you say, there will now be multiple sources
> of truth. If we get different values from all the different streams,
> which one is correct?

Yes, if you resume from same position it will be assumed that source produces sane time mark in the event. But there are a lot of assumptions already, and it's relatively fragile on Windows in other aspects once you start to go off default behavior.

Regarding multiple sources, as I recall it works in general, e.g. you can have two completely separate video rendering branches (topoedit could be used for testing), and it works reliably. Which one is correct we can only test manually, if you return time+10 for one source and time+20 another source and see which one reaches the sink clock state method. What I haven't tried is using two separate source branches and checking if same clock instance is used for both sinks. If it's doing something mad and creates one clock per branch, next question would be what happens to the junction nodes, like transforms with multiple inputs where streams converging (there is an optional IMFClockConsumer thing where you can obviously use only one clock).