[Bug 778830] v4l2dec: Fix race when going from PAUSED to READY

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=778830

--- Comment #3 from Thibault Saunier <tsaunier at gnome.org> ---
> Have you tested some aggressive seeks (where we don't wait for the seek to complete) ? This patch needs work since it make gst_v4l2_video_dec_loop_stopped() and some other code useless. 

Yeah, I ran the validate testsuite with all the scrub seeking tests with that
patch and there seem to be no regression at all (and the testsuite results are
pretty good on the platform, nice work!).

> It's all complicated, and I never found a good solution. What happens is that sometimes, the pad task will stop before gst_v4l2_video_dec_loop() is called. In this case, we ended up with bugs, since the handle_frame() should return FLUSHING.

My patch is about that case in the case of going back to READY, I am not sure
what is your concerns about seeking here.

> I think I was confused when I wrote this. Later Philipp Zabel found that we can read and sync on the pad task state. This is being used in _finish() function. I think the right solution is to completely remove this processing state and to use the pad state instead to detect early flush (while _process() is being called).

That makes sense, will give that a try.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.