On Tue, 17 Dec 2013 21:23:53 +0100, Janne Grunau <[email protected]> wrote: > On 2013-12-16 22:25:53 +0100, Anton Khirnov wrote: > > Otherwise the generic code will unref them, which can then result in > > last_picture_ptr == current_picture_ptr, which causes deadlocks at least > > in rv40. > > I'm undecided, I would say the better fix is to solve the deadlock in > rv40 (or de-mpegenccontextilize rv34) but it looks raher pointless to > allocate dummy frames just to unref them a little later. I can't > remember if I ever knew why the dummyframes are needed but I guess > avoiding checks whether the frames are available in every decoder could > be a reason. ok if you believe this the best solution (ignoring the > de-mpegenccontextilization).
Disregarding the decontextization for now, I believe setting reference for the dummy reference frames is the correct thing to do in any case (since they are after all reference frames). The fact that it happens to fix deadlocks can be viewed as a nice side effect rather than the main goal. And of course decontextizing rv34 should be done at some point. Patches welcome ;) -- Anton Khirnov _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
