At the moment the libcamera monitor and "Device" objects are loaded in the
wireplumber process, but the "Node" objects are loaded in the pipewire daemon.
Due to that, both processes have their own separate `libcamera::CameraManager`
objects. These operate independently.
As a consequence of the above, there is an inherent race condition: when a
new camera appears and the wireplumber process detects it and instructs the
pipewire process to create a "Node" for it, at that point, the camera might
not exist in the `CameraManager` of the "pipewire" process.
This can happen during the initial enumeration as well as hotplug. So load
the nodes locally in the wireplumber process so that all libcamera objects
use the same `CameraManager`, thus eliminating the race condition.
This is required in order to allow plugins to use GL as mincore
is used in Mesas `_eglPointerIsDereferenceable()`.
One example for a client wanting to do so is the in-development
libcamera GPUISP, see https://patchwork.libcamera.org/cover/24183/
(cherry picked from commit pipewire@4796b3fb9524c20ac0f5006143b6a13ee50c01ec)
See pipewire/pipewire!2530
In `wp_interest_event_hook_get_matching_event_types()`, the variable
`res` owns the allocated `GPtrArray`, but there is an early return
in the function. Hitting that will leak the allocation, so use
`g_autoptr` to avoid that.
Fixes: b80a0975c7 ("event-dispatcher: Register hooks for defined events in a hash table")
This simplifies a lot the logic as we don't need to destroy and re-create the
internal BT nodes right away if the loopback nodes were create after them.
We also now listen for changes in the BT profile autoswitch setting. If the
setting is disabled, the source loopback is destroyed. If it is enabled, the
source loopack is created. This makes the setting to take effect immediately,
without needing to disconnect and re-connect the BT device for the setting to
take effect.
As properties are no longer Lua tables, we'll need to make a table copy
for the Json constructor.
Fixes assigning Json.Array to a non-table, and ending up with
"audio.position":"0x7b771267d690" instead of ["FL","FR"] in json output,
which caused BAP device set node to always have mono channels.
Use prefix matching instead of exact matching for media.class when
validating nodes in the set-default command. This allows virtual
nodes (e.g. Audio/Source/Virtual) to be set as default devices,
while still excluding internal nodes.
Fixes#896
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Audio/Sink nodes should only be used as default audio source node type if the
user has explicitly selected it. If the user has not explicitly selected it,
we should always ignore it and instead select the highest priority Audio/Source
node.
Fixes#886
Take the loopback nodes into account also in device Routes.
Some Pulseaudio applications (eg GNOME) determine what to do based on
device ports, so make sure they are consistent with what nodes will be
emitted.
If enabled, this setting will use the same volume levels as the previous
profile. This is useful on some bluetooth devices if the bluetooth profile
audioswitch is enabled.
If the BT profile autoswitch setting is enabled, we also want to create a sink
loopback for SCO-A2DP sink nodes. Since BT nodes are removed and created again
when the profile changes, this avoids confusing some apps making them think
that the BT profile has not changed at all, because the loopback nodes are
always present, even when switching profiles.
This patch improves the BT profile autoswitch logic so that it is simpler and
more robust. Instead of just relying on capture clients that are linked to (or
unlinked from) the BT loopback source node to evaluate whether we have to switch
to a headset profile or not, we now also evaluate the autoswitch every time the
BT loopback source node state changes. This avoids problems with some capture
clients that pause the input stream without closing them.
Apart from this, the patch also fixes some issues with saved profiles if the
user manually switched to a headset profile when no capture streams are present.
The autoswitch logic should restore back the non-headset profile (A2DP) every
time a capture client is disconnected or the BT loopback source node stops
running.
Finally, the 'bluetooth.autoswitch-to-headset-profile' setting will now register
or remove the necessary hooks depending on whether the setting is enabled or
disabled, improving WirePlumber performance if autoswitching is disabled.
When getting the "core" property from a `WpObject`, a strong reference is
returned to the `WpCore` object. This has to be unref-d when not needed
anymore. `wp_modem_manager_enable()` fails to do so, so fix it.
Fixes: 2794764d5a ("m-modem-manager: add module for tracking status of voice calls")
Add immediate linking for common cases to reduce latency when new
audio/video streams are added. The full rescan mechanism is preserved
and still triggers for all session-item-added events via the
rescan-trigger hook.
Recompiling with the SPA_AUDIO_MAX_CHANNELS=128u defined, will create
larger audio_info structures that can hold more channels. Because
PipeWire is now using 128 channels, do the same in wireplumber so that
the generated PortConfig param can hold all channel positions.
See pipewire#4995
And include the error message while we're at it.
Fixes: 2794764d5a (m-modem-manager: add module for tracking status of voice calls)
Signed-off-by: Evangelos Ribeiro Tzaras <devrtz@fortysixandtwo.eu>
We segfault otherwise on error:
Thread 1 "wireplumber" received signal SIGSEGV, Segmentation fault.
g_type_check_instance_is_fundamentally_a (type_instance=type_instance@entry=0x7fffec008120,
fundamental_type=fundamental_type@entry=0x50 [GObject]) at ../../../gobject/gtype.c:3917
warning: 3917 ../../../gobject/gtype.c: No such file or directory
(gdb) bt
#0 g_type_check_instance_is_fundamentally_a
(type_instance=type_instance@entry=0x7fffec008120, fundamental_type=fundamental_type@entry=0x50 [GObject])
at ../../../gobject/gtype.c:3917
#1 0x00007ffff7eafe3d in g_object_unref (_object=0x7fffec008120) at ../../../gobject/gobject.c:4743
#2 0x00007ffff4784ec1 in list_calls_done (obj=<optimized out>, res=<optimized out>, data=0x5555556ab1b0)
at ../modules/module-modem-manager.c:209
Fixes: 2794764d5a (m-modem-manager: add module for tracking status of voice calls)
Signed-off-by: Evangelos Ribeiro Tzaras <devrtz@fortysixandtwo.eu>
This avoids sorting them constantly when collecting them. If a hook has a
circular dependency, a warning will be logged and the hook won't be registered.
See #824
