Add new context config clock.power-of-two-quantum to make it possible to
not round quantum to closest lower power of two. This makes it possible
to match the quantum of a source node with the quantum of a client
node.
Allows monitors (eg. bluez5) free any sm_objects they hold in
session_destroy. In session_shutdown, destroy such objects before the
event, and free them only afterward.
Fixes double-free in bluez monitor.
WE can only delay the start for drivers that are driving. Remote
nodes are driving themselves but are not drivers so they need to
start early.
See #879
A node starts out as passive but becomes non-passive when there
is a non-passive link to an active node. If there is a link to
an inactive node, the node remains passive.
This fixes suspend of drivers in all cases.
Fixes#898
Don't always try to restore the saved profile when it's available
but only do this the first time. From then on, try to follow the
best profile when it changes.
If a driver node is linked with a non-passive link to itself, it
should also become running.
One example of this is the Midi bridge, that needs to start when you
link one device to another. The devices are all on the same node
so linking devices would result in loopback connections.
libcamera + bluez5 devices should be freed/unloaded via the sm_object
free callback, similarly as in alsa and v4l2 monitors. This ensures
they are run at session_shutdown.
To resolve monitor and policy core global ids racing with each other,
use separate registry event handlers for both cores. Each handles only
their own objects, determined by where the object handle was created.
Postpone handling of policy core new global events after monitor sync,
which orders them after the corresponding monitor proxy and registry
events. Monitor core is then more up-to-date, so we resolve id clashes
in favor of monitor globals, which avoids duplicate objects.
Fix use-after-free by tracking whether a monitor holds references to
sm_object. Keep also objects pending for id in a list, so that they can
be cleaned up on session_shutdown (monitors may leak objects at
shutdown, because spa objectinfo events won't be handled then).
Caveats:
Zombie objects may still created if policy core is late by several
events, but in those cases the corresponding remove events are already
in the queue.
Also, there's a (theoretical) possibility that pw_registry_bind will
bind the wrong object, if the registry event is handled too late and an
id is reused by the server.
For details, see reverted 77e4fdb1e4
for which this is a another approach.
This simplifies the code and also avoids a warning with _FORTIFY_SOURCE:
../pipewire/src/modules/module-protocol-pulse/pulse-server.c: In function ‘create_pid_file’:
../pipewire/src/modules/module-protocol-pulse/pulse-server.c:6028:2: warning: ignoring return value of ‘write’ declared with attribute ‘warn_unused_result’ [-Wunused-result]
6028 | write(fd, pid_str, strlen(pid_str) + 1);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Make arrays from config sections that should really be an array.
Having the module name as the object key technically makes it
impossible to load the same module twice because the key can only
be once in the object.
The same applies to the context.objects and context.exec sections.
This also makes it somewhat easier to parse the config..
Because we set the default sink/source before calling the metadata,
we don't send out updates about the new sink and this confuses
pavucontrol.
Instead, use the new async reply to reply once the new metadata has
been configured and the defaults have been updated.
We only want to restore the saved profile when it is available, when
it is not available and there was no change in best profile, leave
the profile as it is.
Keep track of the best profile. Only try to switch away from the
saved profile when something changed and the saved profile is not
available.
This makes it possible to select an unavailable profile but when
something is changed, it will switch away from it.
A node.target property of -1 also means to follow the default
sink/source. The ALSA plugin might use this to select the
target node and would otherwise not move to the new default sink.
Only start the driver node after we have added ourselves to the
driver node or else we might start too soon, before there is anything
to schedule and we cause driver underruns.
Only activate the links when the node is added to a driver. Otherwise
the driver will expect us to decrement the activation counters for the
links but that won't happen because the node will not be processed
yet.
Instead we activate the links that were not activated yet when the
node is activated.
This improves scheduling for newly added links and nodes.
When we were busy with the graph recalculation and something changed,
restart the graph state change to get the new change applied.
For example, when we start a node and it has a max-quantum set, we
need to restart our calculation with the new quantum limits.
Use the manager sync method to wait until all pending queries are
finished. It is the manager that will query the new parameters so
we need to wait until that completes.
See #868
Trigger a roundtrip before we send the reply to the client for many
methods like set_volume/set_mute etc.
this ensure that we get the updated information from the server before
we reply so that a subsequent query returns the up-to-data information.
Fixes#868
