Fallback in video_device_monitor_impl.cpp dosn't work.

Udev isn't supported inside a Flatpak sandbox and judging by the code inside video_device_monitor_impl.cpp it should automatically fall back to enumerating /dev/video# devices. For some reason this never gets executed and patching out the udev related code triggers some infinite loop. Since I'm not a c++ expert my patch could be at fault as well. Last where are the JAMI_ERR() calls going, I don't see anything in stdout?

jami-daemon_no-udev.patch

Yeah I never got the video working inside flatpak Note that a snap is available and working.

Anyway with your patch, what do you see?

Also how do you launch dring? (dring -cd?)

Can you try with:

diff --git a/src/media/video/v4l2/video_device_monitor_impl.cpp b/src/media/video/v4l2/video_device_monitor_impl.cpp
index c6a4a2e9f..a8139c8f5 100644
--- a/src/media/video/v4l2/video_device_monitor_impl.cpp
+++ b/src/media/video/v4l2/video_device_monitor_impl.cpp
@@ -36,6 +36,7 @@
 #include "../video_device_monitor.h"
 #include "logger.h"
 #include "noncopyable.h"
+#include <fstream>
 
 extern "C" {
 #include <fcntl.h>
@@ -177,6 +178,10 @@ udev_failed:
     for (int idx = 0;; ++idx) {
         std::stringstream ss;
         ss << "/dev/video" << idx;
+        auto stream = std::ifstream(ss.str());
+        if (!stream.good())
+            return;
+        stream.close();
         try {
             monitor_->addDevice(ss.str());
         } catch (const std::runtime_error& e) {

The main diff is that we check that /dev/videoX is present and if not leave the loop. addDevice changed a bit since some time and the loop can be infinite if nothing throw

My patch stalls jami-daemon and i see nothing. If I add yours jami-daemon works and lists my camera but that's all. The the preview window stays black and the little control LED next to the camera stays off as well.

dring gets started by jami-gnome over dbus.

When trying to record a video: ** (jami-gnome:2): WARNING **: 18:05:42.052: set_state: failed to start recording, wait preview

or video call: could not open shm area "Jami Daemon_shm_2_0" , shm_open failed: File or directory not found

can you start the daemon manually and give logs?

Also in the flatpak is the video devices available?

Log: jami-daemon.log

v4l2-ctl shows the device as well as lsusb

looks like it's a ffmpeg issue judging by Cannot find format "video4linux2"

Video calls work after adding a custom ffmpeg build with v4l support.

The only confusing thing is that if starting dring manually the video preview in settings and video calls work but starting dring via dbus breaks all previews. Recording video in chat or video calls still work but you don't see yourself.

mmh if you start to use a custom ffmpeg, imho you can just build the one in daemon/contrib/ as it will also support REMB feedback (adaptative bitrate)

For the black renderer, I have 2 questions:

1. What tuple daemon/lrc/client-gnome are you using? There was some changes in the API recently for renderers so I recommend master/master/master

2. In a video call, do you see the other side? If so, I think it's related to my first question

I'm now on master/master/master and the issue still persists. Works with manually started dring and black previews when using dbus.

On my phone I see everything and only a black window on my desktop.

flatpak build is https://github.com/flathub/net.jami.Jami/pull/25

dbus-jami-gnome.log

manually-jami-daemon.log

so the daemon is working and this is at least a nice part.

I'm not sure that the patch for v4l2/video_device_monitor_impl.cpp is necessary anymore since you changed ffmpeg with v4l2 support. I still recommend the one in the daemon/contrib directory

For the logs, it seems "ok" except libcanberra is missing, but it's for sound in notifications so not related. And gnutls should be 3.6.15 like the contrib directory, but not related too.

Commits seems ok too.

So there is 2 possibilities we can dig:

1. If you run dbus-monitor | grep cx.ring.Ring.VideoManager you should see camera events (startedDecoding, stoppedDecoding, startCamera, stopCamera)
2. It is certainly a flatpak issue (as the same code show the camera on other packages). And a permission may miss, journalctl/strace can give you inputs for opened files that can't be opened?

cc @abandali this topic may interest you

Another way to check is to add logs in lrc/src/shmrenderer.cpp to check if frames from the shared memory can be retrieven by the client (but first thing to check is if flatpak doesn't fail to get some files)

I tried --filesystem=host --filesystem=home --socket=session-bus and it still has the shm errors. These are the most open permissions.

Removing your patch makes dring try to add infinite /dev/video devices.

Using a wrapper to start dring and jami-gnome separately fixes the preview issue.

ooh, that's weird

I may have a solution without the wrapper. We can do as I do for appimage, use libwrap, this will avoid dbus

https://review.jami.net/c/ring-project/+/16627

  appimage='-DENABLE_LIBWRAP=true'

  static='-DENABLE_STATIC=true'

Adding -DENABLE_LIBWRAP=true works as well

Cool. Imho it's the way to go for flatpak. I don't think we want to run the daemon separately here and the flatpak should be used to launch the whole thing (daemon+client together). Moreover, for the logs, all will be done.

I am currently doing the patch for the daemon. Will ping you for testing and once review is done, I think we may close this issue unless you see something else?

closed

fyi https://review.jami.net/c/ring-daemon/+/16764

Thanks, it works!

mentioned in commit 3eede614