On Wed, Apr 27, 2022 at 10:48:42PM +0200, Caspar Schutijser wrote: > Hi, > > On Thu, Mar 24, 2022 at 07:11:42AM +0100, Alexandre Ratchov wrote: > > Most audio/video players do a stop/start cycle whenever the play > > position is changed, track is changed, etc. Currently, stopping drains > > the play buffer, which by default is very large (to workaround very > > long kernel non-preemptive code-paths). This makes player controls > > sluggish. > > > > This diff adds a new sio_flush() function to workaround the jumbo > > buffer sizes: it stops playback immediately, discarding buffered > > data. Basically it's the same as sio_stop() but doesn't wait. The plan > > is to make players use it. > > > > In the network protocol, sio_flush() is implemented by adding a flag > > to the message corresponding to sio_stop(). Old sndiod servers ignore > > it and just work with new libraries. New sndiod servers see that the > > flag is not set by old libraries and properly drain the play buffer. > > > > Tested with mplayer, mpv and audacious, if we go this way other ports > > will follow. > > I tested this with Audacious and I like the new behavior a lot! When I > change the play position, stop playback or change the track it indeed > responds much faster. > > When pausing playback however, the buffer is still being drained which > suprised me a little bit. However, after some printf-debugging I > confirmed my suspicion that Audacious does not call SndioPlugin::flush() > in that case, so this patch (or sndio in general) doesn't have anything > to do with that. > > I can't really comment on the diff itself, although that makes sense > to me. >
thank you for testing this. I retested multimedia/mpv and x11/mplayer as well; besides the need for an extra REVISION bump, my previous diff still applies. OK?
