Direct `--player pipewire` output support for `snapclient`?

[k8ieone]

So PipeWire seems to be slowly getting more popular. And I get why, I mean it can interface with ALSA and PulseAudio applications while promising to be a low-level and low-latency API as well.

I've never had to deal with programming any kind of audio-related program, so I have only a very limited understanding of audio protocols, so this might sound stupid.
Would adding a pipewire option to --player decrease latency as opposed to running snapclient with --player alsa/pulse and having PipeWire "translate" it? If there was a latency decrease, would it be significant enough to justify the effort needed to code support for another audio API?

[badaix]

PipeWire is an audio server, just like PulseAudio. Under the hood both of them are using ALSA (on Linux), because ALSA provides an interface to the Linux kernel audio driver. If you aim for low latency, you should use the alsa backend.
In it's core ALSA provides an exclusive access to the audio device, meaning that only one process can play audio at a time, alsa provides also some higher level virtual mix interfaces, but usually something like PulseAudio is used for non-exclusive access.
Because many applications do not natively support PulseAudio or PipeWire, they provide a virtual default ALSA device that will not interface with the audio device, but forward the audio stream to PA/PW and is getting processed there (effects, volume, etc) and mixed into the output stream that is fed to the "physical" Alsa device.
So on a system with PulseAudio you have several alsa devices (snapclient -l):

0: default
Playback/recording through the PulseAudio sound server
...
31: hw:CARD=Generic_1,DEV=0
HD-Audio Generic, ALC256 Analog
Direct hardware device without any conversions

If you use --player alsa -s 0 the audio will go through this pipeline:

Snapclient => ALSA => "virtual PA device" => PA => ALSA => device 31

If you use --player alsa -s 31 it is

Snapclient => ALSA => device 31

This will yield lower latency, but on your desktop system with PA you will not hear any other audio or Snapclient will complain that it cannot connect to the device, if audio is already playing through PA (PA releases the ALSA device, just as Snapclient does when being idle).

If you use --player pulse:

Snapclient => PA => ALSA => device 31

This will yield lower latency than --player alsa -s 0 and better error handling, control and jitter and other processes will be able to play audio, but it's higher latency and more jitter than --player alsa -s 31, but still preferred for a desktop system (it will also provide it's own volume control for PA).

All this will also apply for PipeWire. So, on systems running PipeWire instead of PulseAudio, it would make sense to have native support for it. For headless clients, there is the strong recommendation to use ALSA.

Edit: I would guess that PipeWire's PulseAudio server emulation should be close to PipeWire's native client library (performance and feature wise).

[k8ieone]

Very nice explanation, thanks. I understand ALSA's exclusive access limitation (thus the reason for PA's existence).

Just out of curiosity, is PipeWire's PA emulation similar to Pulse's ALSA? What I mean is if it also creates a "fake device" (in Pulse's case I believe that would be called an output sink?) that programs can talk to, like Pulse does with ALSA?
Or is PipeWire's Pulse emulation just an alternative implementation of the PA protocol?

snapclient => PA => PW => ALSA => device

or like this:

snapclient => PW (PA emu mode) => ALSA => device

If it's a re-implementation of the PulseAudio protocol, is there any point in making programs interface directly with PipeWire, using their protocol instead of talking to it like they would with PulseAudio?
I'd guess for having a program work universally, it should use ALSA, because it can either work with devices directly or go through either PipeWire or PulseAudio.

[k8ieone]

Wow, this turned into a QnA session... You don't really have to answer if you don't want to.

TLDR: For snapclient running on a PipeWire system, it's best to use --player pulse and it's not really worth it to implement PipeWire's native libraries. Correct?

[badaix]

Well, I have no idea about PipeWire, this is just my 5min party knowledge, I've gathered from reading their GitLab Readme.
I think the idea behind PipeWire is that it not only focuses on Audio, but also on Video, and maybe it's better or more modern in some regards as PulseAudio, or the authors just don't like Poettering 😉
PulseAudio and PipeWire are running as server and they offer client libs to connect to it and play audio through it. If I would invent a new audio server (well, actually Snapcast is an audio server), I would also probably make the server compatible to the PulseAudio client libs, so that it can be a drop in replacement, while the PipeWire client lib might give better control or might have more features.
Bottom-line: I don't think that a --player pipewire which uses the native PipeWire client lib would give better results, since it does quite basic audio playback, but depends on reliable playback latency estimations, which would hopefully work.

[ssieb]

PipeWire is intended to combine the features of both pulseaudio and Jack. It is supported by Poettering, there's no competition involved. It is a drop-in replacement for pulseaudio, but in general, using native interfaces is better than using the compatibility ones.

[badaix]

@ssieb thanks for the info, do you have an idea on how widespread this is used, i.e. how man distributions use it as their default audio server
@satcom886 I've created a "Feature requests" project to keep this issue tracker clean and to not drop this feature request, I have linked it there and will close it here.

[ssieb]

I don't think any do yet. I just know that Fedora (that I use), is intending it to be the default for the release that is being worked on right now.
https://fedoraproject.org/wiki/Changes/DefaultPipeWire

[k8ieone]

Seems like PipeWire's devs actually recommend sticking with ALSA/PulseAudio libraries for now.
https://gitlab.freedesktop.org/pipewire/pipewire/-/wikis/FAQ#what-audio-api-do-you-recommend-to-use

[PureTryOut]

Fedora will switch to it with release 34, but so far that is the only distribution as far as I know. A lot of users are switching already however due to PulseEffects having dropped support for PulseAudio and moving fully to Pipewire. It's hard to give an exact number, or even an estimate, of people using Pipewire right now but it's steadily increasing.

I can say I would very much like direct Pipewire support 🤗

[rwjack]

Is there a way to merge regular Desktop audio with SnapClient?

I'm Running popOS with pipewire-alsa

[deisi]

@ssieb thanks for the info, do you have an idea on how widespread this is used, i.e. how man distributions use it as their default audio server

By now Fedora (>34) and Ubuntu (>22.10) use pipewire by default:

• https://www.omgubuntu.co.uk/2022/05/ubuntu-22-10-makes-pipewire-default
• https://fedoramagazine.org/pipewire-the-new-audio-and-video-daemon-in-fedora-linux-34/

As a roule of thumb, if the disto uses wayland, chance are it is using pipewire as well. In any case its becoming increasingly relevant.