Bug 296007 – REGRESSION(292932@main): [GTK][WPE] Output audio device enumeration does not work

• Component: New Bugs
• Status: RESOLVED FIXED

[Adrian Perez]

Using navigator.mediaDevices.enumerateDevices() no longer works in ToT. Bisecting from a good commit landed on 292932@main being the culprit.

Steps to reproduce

  % run-minibrowser --wpe -- \
    --features='ExposeSpeakers,ExposeSpeakersWithoutMicrophone,PerElementSpeakerSelection,-SpeakerSelectionRequiresUserGesture,LogsPageMessag
esToSystemConsole' \
    https://people.igalia.com/aperez/joplin.html


Expected outcome:

  Audio output devices are enumerated, logged accordingly, and a checkbox is added to the page for each of them.

Actual outcome:

  The promise returned by enumerateDevices() never gets resolved.

[Philippe Normand]

Works fine here:

Accepting WebKitUserMediaPermissionRequest request
https://people.igalia.com/aperez/joplin.html:24:14: CONSOLE LOG getUserMedia success, fetching devices, stream: [object MediaStream]
https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG Audio device enumeration started...
https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG stopping audio track "934ed5fc-549e-444c-a299-47a02ec0a2ca" (Built-in Audio Analog Stereo)
https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG Device "6F63BCAE32663CBF0A043CAFD1B013139CE7534C" skipped, type is "audioinput"
https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG Device "default" added: "Default - Built-in Audio Analog Stereo"
https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG Device "A94CCA91C199013F31E1F1FA0267CAF86870F53B" added: "Built-in Audio Analog Stereo"

Can you share a GST_DEBUG="3,webkit*:6" log please?

[Philippe Normand]

I guess you don't have desktop portals on the device, right?

[Adrian Perez]

(In reply to Philippe Normand from comment #2)
> I guess you don't have desktop portals on the device, right?

That's right, the device for which I first implemented device enumeration/setSinkId does not have portals; and not even PipeWire -- it uses ALSA directly.

I forgot to mention above in the issue description that the command used to reproduce needs to be used with the Flatpak SDK. When I noticed that using Flatpak would hit the issue then I continued using it to do the bisection faster.

[Adrian Perez]

Created attachment 476071 [details]
GStreamer log with GST_DEBUG='3,webkit*:6'

Here's the log, running with the following command using the Flatpak SDK, with a build done on commit 297517@main:

   GST_DEBUG_FILE="$(pwd)/gstreamer.log" \
   GST_DEBUG='3,webkit*:6' run-minibrowser --wpe -- \
   --features='ExposeSpeakers,ExposeSpeakersWithoutMicrophone,PerElementSpeakerSelection,-SpeakerSelectionRequiresUserGesture,LogsPageMessagesToSystemConsole' \
   https://people.igalia.com/aperez/joplin.html

The console output is:

   SDK origin: webkit
   Using default MiniBrowser
   Mesa: The provided filesystem timestamp for the cache is bogus! Disabling On-disk cache.
   Accepting WebKitUserMediaPermissionRequest request
   https://people.igalia.com/aperez/joplin.html:24:14: CONSOLE LOG getUserMedia success, fetching devices, stream: [object MediaStream]
   https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG Audio device enumeration started...
   https://people.igalia.com/aperez/joplin.html:32:14: CONSOLE LOG stopping audio track "0f1105c5-d663-435f-9f26-75066da9c890" (Jabra Link 380 Mono)
   ^C

[Adrian Perez]

Checking quickly the log, it seems that at least the capture device manager picks the outputs (the mono devices are inputs, the stereo ones are two outputs that should be enumerated as spekers):

   0:00:01.846763888    37 0x5a27678d8290 INFO    webkitcapturedevicemanager GStreamerCaptureDeviceManager.cpp:248:addDevice:<pipewiredevice1> Registering non-default device Jabra Link 380 Analog Stereo
   0:00:01.846798102    37 0x5a27678d8290 INFO    webkitcapturedevicemanager GStreamerCaptureDeviceManager.cpp:248:addDevice:<pipewiredevice2> Registering non-default device Jabra Link 380 Mono
   0:00:01.846817389    37 0x5a27678d8290 INFO    webkitcapturedevicemanager GStreamerCaptureDeviceManager.cpp:248:addDevice:<pipewiredevice3> Registering non-default device KM_B2 Digital Audio Analog Stereo
   0:00:01.846835182    37 0x5a27678d8290 INFO    webkitcapturedevicemanager GStreamerCaptureDeviceManager.cpp:248:addDevice:<pipewiredevice4> Registering non-default device KM_B2 Digital Audio Mono

The bulk of the log seems to be related to the autoplayed audio element from the test page. I could remove the “autoplay” and get another log with different options if that would be helpful =]

[Philippe Normand]

> PipeWireCaptureDeviceManager.cpp:65:computeCaptureDevices: Checking with Camera portal

Somehow the DesktopPortal is created and is valid, there's no "Portal not present or has no camera" log message... So portal->accessCamera() is called and hangs there, apparently...

I don't understand that you have no portal running, but still the dbus proxy seems created and connected to /org/freedesktop/portal/desktop? 
Maybe we're mis-using g_dbus_proxy_new_sync()...

Can you capture a dbus log with dbus-monitor?

[Philippe Normand]

Also it's worth testing with an early return from PipeWireCaptureDeviceManager::computeCaptureDevices() (call `callback()` before returning).

[Philippe Normand]

Pull request: https://github.com/WebKit/WebKit/pull/48241

[Adrian Perez]

(In reply to Philippe Normand from comment #8)
> Pull request: https://github.com/WebKit/WebKit/pull/48241

Indeed, with this applied I get device enumeration working again as expected.

[EWS]

Committed 297669@main (04d2f39a0b79): <https://commits.webkit.org/297669@main>

Reviewed commits have been landed. Closing PR #48241 and removing active labels.

[Radar WebKit Bug Importer]

<rdar://problem/156295253>