[Hengqian Ling (linghengqian)]

• Fixes use Pipewire instead of PA #290 .
• Replace PulseAudio with PipeWire/WirePlumber audio stack in WSLg.

[Hengqian Ling (linghengqian)]

@microsoft-github-policy-service agree

[Ben Hillis (benhillis)]

Triage note (2026-05): Thanks for the substantial work — replacing PulseAudio with PipeWire/WirePlumber is squarely on the roadmap (tracked in #290, with 40+ reactions) and we want to land this. A few things will hold the merge button:

1. We need to validate audio behavior end-to-end against current RDP audio playback and capture paths. The pipewire-wslg-rdp.patch and pipewire-rdp-module.c carry the equivalent of the existing PulseAudio RDP module — verifying parity (sample rates, latency, capture devices, mute/volume integration with Windows) is the bulk of the review.
2. cgmanifest.json will need entries for PipeWire / WirePlumber so Component Governance is happy.
3. An internal buddy build through the full Azure Pipelines flow before merge.

Not asking for changes yet — just setting expectations. If you want to chip away at (1) and (2) preemptively that would help.

[Hengqian Ling (linghengqian)]

I'm glad Microsoft is willing to take on this work, especially since I don't actually have an Azure account to test the corresponding CI files. Should I close this PR?

[Rebecca Turner (iarna)]

especially since I don't actually have an Azure account to test the corresponding CI files

They were suggesting that a microsofty would need to do the full run through, which is why they suggested you could, if you wanted, look at 1 and 2.

They seem to be saying they're gonna do the work to land this one way or the other, so that would imply that no, you should not close it.

(Obviously, as neither a member of this project, nor an MS employee, I'm not speaking for anyone here, just hopefully clarifying things.)

[Hengqian Ling (linghengqian)]

I am not sure whether I should resolve the merge conflicts; even if I did, I wouldn't be able to register for an account on https://www.azure.cn/ as an individual to test the CI, as accounts there require corporate registration.

[Copilot]

🟡 Changes recommended

Native modules cannot currently load through pulse.cmd, and several real-time socket and startup issues can prevent or stall audio.

14 open findings

• Use PIPEWIRE_LOG instead of unsupported --log · New
• Use PIPEWIRE_LOG for pipewire-pulse logging · New
• Load native RDP modules through context.modules · New
• Suppress SIGPIPE during socket writes · New
• Disconnect only owned PipeWire core connections · New
• Remove blocking I/O from the real-time callback · New
• Use nonblocking framed reads in the real-time callback · New
• Gate WirePlumber startup on PipeWire readiness · New
• Set sink properties before connecting the stream · New
• Use requested buffer capacity instead of stale chunk size · New
• Clear output buffer metadata after socket errors · New
• Set source properties before connecting the stream · New
• Document the actual default PipeWire log file · New
• Correct WSLg Mirror links and patch provenance · New

What changed in this PR

Replaces WSLg’s PulseAudio implementation with PipeWire, WirePlumber, and PulseAudio compatibility support, addressing #290.

Changes:

• Builds and launches the PipeWire audio stack.
• Adds custom PipeWire RDP sink/source modules and configuration.
• Updates build metadata and documentation.

File	Description
WSLGd/​main.cpp	Launches and monitors the new audio stack.
samples/​container/​Containers.md	Updates container audio guidance.
README.md	Documents PipeWire architecture and verification.
Microsoft.WSLg.nuspec	Updates package description.
Dockerfile	Builds and packages PipeWire and WirePlumber.
CONTRIBUTING.md	Updates build prerequisites and arguments.
config/​pipewire.conf	Configures the PipeWire daemon.
config/​pipewire-wslg-rdp.patch	Adds RDP modules to PipeWire’s build.
config/​pipewire-rdp-module.c	Implements RDP audio bridging.
config/​pipewire-pulse.conf	Configures PulseAudio compatibility.
config/​BUILD.md	Updates manual build instructions.
cgmanifest.json	Registers new dependencies.
build-and-export.sh	Supplies new dependency revisions.
.github/​ISSUE_TEMPLATE/​bug_report.yml	Updates requested audio logs.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

The RDP module contains out-of-bounds and real-time-thread blocking risks and does not preserve transport latency reporting.

4 open findings

• Blocking socket I/O in PipeWire real-time callback · New
• Chunk size can read past buffer end · New
• RDP transport latency is not reported to PipeWire · New
• realloc in real-time callback can stall audio · New

14 resolved since last review

• Use nonblocking framed reads in the real-time callback
• Remove blocking I/O from the real-time callback
• Disconnect only owned PipeWire core connections
• Suppress SIGPIPE during socket writes
• Load native RDP modules through context.modules
• Use PIPEWIRE_LOG for pipewire-pulse logging
• Use PIPEWIRE_LOG instead of unsupported --log
• Set source properties before connecting the stream
• Clear output buffer metadata after socket errors
• Use requested buffer capacity instead of stale chunk size
• Set sink properties before connecting the stream
• Gate WirePlumber startup on PipeWire readiness
• Correct WSLg Mirror links and patch provenance
• Document the actual default PipeWire log file

🧠 Review effort: Balanced

[Copilot]

🟡 Changes recommended

The RDP modules contain latency, concurrency, real-time processing, and lifecycle defects that can break or degrade audio.

7 open findings

• Race from mixed atomic and non-atomic transport latency access · New
• Blocking socket operations run in the real-time callback · New
• pthread_create failure incorrectly reports stale errno · New
• Build embeds unmanifested Lua instead of using system Lua · New
• Source socket remains open when the stream is paused or suspended · New
• Docker disables the X11 bell module and silently regresses audio · New
• wsl --system command incorrectly includes a distro name · New

4 resolved since last review

• Chunk size can read past buffer end
• Blocking socket I/O in PipeWire real-time callback
• realloc in real-time callback can stall audio
• RDP transport latency is not reported to PipeWire

🧠 Review effort: Balanced

[Copilot]

🟡 Changes recommended

The RDP source has concurrency and buffer-accounting defects, and both audio modules fail to recover after PipeWire restarts.

4 open findings

• Reconnect sink after monitored PipeWire restart · New
• Reconnect source after PipeWire restart · New
• Set PipeWire buffer size to queued frame count · New
• Clear stale suspend request when handling Start · New

7 resolved since last review

• pthread_create failure incorrectly reports stale errno
• Blocking socket operations run in the real-time callback
• Race from mixed atomic and non-atomic transport latency access
• Docker disables the X11 bell module and silently regresses audio
• Source socket remains open when the stream is paused or suspended
• Build embeds unmanifested Lua instead of using system Lua
• wsl --system command incorrectly includes a distro name

🧠 Review effort: Balanced

[Copilot]

🟡 Changes recommended

The RDP module has incorrect latency reporting, unsafe core reconnection lifecycle handling, and possible microphone frame corruption after overflow.

3 open findings

• Core proxy destruction leaves stale listeners and pointers · New
• Transport latency is inflated by additive unit misreporting · New
• Overflow draining can misalign S16 audio frames · New

4 resolved since last review

• Clear stale suspend request when handling Start
• Set PipeWire buffer size to queued frame count
• Reconnect source after PipeWire restart
• Reconnect sink after monitored PipeWire restart

🧠 Review effort: Balanced

[Copilot]

🔵 Needs a closer look

Latency publication, configurable RDP socket paths, and native socket access have unresolved functional regressions.

0 open findings

3 resolved since last review

• Core proxy destruction leaves stale listeners and pointers
• Overflow draining can misalign S16 audio frames
• Transport latency is inflated by additive unit misreporting

Previously missed (4)

In code that hasn't changed since last review

Preserve PULSE_AUDIO_RDP_SINK endpoint override

config/​pipewire-pulse.conf:36

This hardcoded path bypasses the existing PULSE_AUDIO_RDP_SINK override set by WSLGd and configurable through .wslgconfig; the former PulseAudio RDP module read that environment variable. Private/test deployments that redirect Weston's sink socket will now keep connecting to /mnt/wslg/PulseAudioRDPSink. Preserve the environment fallback in the new module and avoid overriding it here.

Preserve PULSE_AUDIO_RDP_SOURCE endpoint override

config/​pipewire-pulse.conf:47

This likewise bypasses the existing PULSE_AUDIO_RDP_SOURCE override set by WSLGd and configurable through .wslgconfig; the former source module consumed that variable. Any deployment using a redirected microphone socket will silently connect to the default path instead. Preserve the environment fallback in the module rather than fixing the endpoint in this config.

Defer stream parameter updates to the main loop

config/​pipewire-rdp-module.c:616

pw_stream_update_params() is called from the RT/data-loop process callback, but PipeWire 1.4.10 enforces that this API runs on the stream's main loop (ensure_loop(impl->main_loop, return -EIO)). The update therefore fails, and published_seq is still advanced, so the measured RDP transport latency is never published or retried. Defer the parameter update to the main-loop timer and only mark the sequence published after a successful call.

Configure native PipeWire socket access permissions

config/​pipewire.conf:27

The native pipewire-0 socket is created without explicit permissions, so protocol-native leaves its mode to the process umask; unlike the Pulse socket (which PipeWire chmods to 0777), it is not connectable by user-distro accounts whose UID differs from the fixed wslg UID 1000. That prevents the native PipeWire/JACK/Flatpak use case motivating this migration for such accounts. Configure socket ownership/mode and the matching access policy explicitly for projected user-distro access.

🧠 Review effort: Balanced

[Hengqian Ling (linghengqian)]

Needs a closer look

• I’ve reviewed it and don’t see any issues; I tested it locally using that Dockerfile in rootless Docker.