Re: [Call for testing] sound: Implement /dev/dsp as a router device

Jakob Alvermark wrote:
> On 2025-04-08 00:23, Christos Margiolis wrote:
> > Hello Jakob,
> > 
> > Jakob Alvermark wrote:
> > > Hi,
> > > 
> > > 
> > > This is cool.
> > > 
> > > I did some quick testing.
> > > 
> > > Playing some audio and plugging in my USB-C headset the audio switches over
> > > to the headset and continues playing in the headset.
> > > 
> > > However, unplugging the headset does not seem to work fully. If I play
> > > something in chromium (a Youtube video for example), it stops.
> > See what I write below, it could be a similar issue. Also, does this
> > happen on other browsers, say, firefox?
> 
> Interestingly, with firefox, the audio does not switch over to the headset.
> 
> I tried this:
> 
> - Start playing audio in firefox and start playing some audio in chromium.
> 
> - Plug in USB-C headphones.
> 
> - The audio from firefox still plays through the laptop speaker, but the
> chromium audio switches over to the headphones.
> 
> - Unplug headphones.
> 
> - Firefox audio still plays through laptop speakers as if nothing happened,
> chromium audio stops.

We can test a few things:

- Can you share your /dev/sndstat with hw.snd.verbose=2 when you
  encounter the errors you mentioned?
- Do you get the expected result if you do _not_ (un)plug anything
  during IO, but instead just switch between devices with "mixer -d
  pcmX", where X is the device's number?

Also it's important to note that plugging a new device does not mean
sound(4) will automatically switch to it. It also doesn't mean that
unplugging a device will necessarily switch to the device you want. The
driver does include a heuristic for choosing the default device, though.

> 
> > > I tried playing an mp3 file using audio/mpg123, and when unplugging the
> > > headset I get this error:
> > > 
> > > [src/libout123/libout123.c:out123_play():746] error: Error in writing audio,
> > > wrote only -1 of 2304 (Invalid argument?)!
> > > main: [src/mpg123.c:play_frame():857] error: Deep trouble! Cannot flush to
> > > my output anymore!
> > I am not familiar with audio/mpg123, but the error message suggests that
> > the program exits upon a write() error. That is not really an issue of
> > sound(4), but that audio/mpg123 exits on the first write() error. mpv
> > for example, or mplayer as you said, do not have this problem.
> > 
> > My first guess would be that the error is generated during the
> > hot-unpluging, because in the time window between the device detach and
> > the switch to the new default device, write() will have failed since the
> > device stopped doing IO at this point. That is not wrong behavior on
> > sound(4)'s side though.
> > 
> > > On the other hand, multimedia/mplayer does work, switching seamlessly
> > > between the laptop speaker and headset.
> > Just to make sure, the switching is successful even during IO, right?
> Yes, with mplayer, switching works fine, both plugging and unplugging, while
> playing an mp3 file.
> > > Jakob
> > > 
> > > On 2025-04-06 21:08, Christos Margiolis wrote:
> > > > I've been working on a patch [1] to re-implement /dev/dsp as a
> > > > router/virtual device. The commit message in the review explains all the
> > > > high-level details and reasons behind the patch. This is the first of a
> > > > series of patches I'm working on.
> > > > 
> > > > For regular users, the main functional difference is that with this
> > > > patch, /dev/dsp is an actual router device (like when using
> > > > virtual_oss), which means that we have hot-swapping (i.e., changing the
> > > > default device and routing audio to it during IO without needing to
> > > > close and re-open /dev/dsp) built into sound(4).
> > > > 
> > > > Feel free to test the patch and please let me know if there's anything
> > > > that breaks or could be improved. You can play around with it just by
> > > > recording/playing audio as you always do, try some hot-swaps,
> > > > hot-(un)plugs, and anything that can stress test it. Just make sure you
> > > > open /dev/dsp, which should be the default for most applications anyway.
> > > > 
> > > > Christos
> > > > 
> > > > [1] https://reviews.freebsd.org/D49216
> > Christos

Christos