Regular filters should be treated as regular stream / device nodes. Note that
the filter utils API is meant to be used using the direction of the main smart
filter nodes. This is why we use the target direction instead of the actual
stream direction to check if the stream session item is smart or not.
This should never happen, but there are odd cases with broken
configuration where such a device may appear and the least we can do
is not crash, at least not when the device.name is only used for
debug messages. In state-profile and other places where the name
really matters, we actually have checks in place.
Fixes: #674
We always have to destroy the restore timeout source when we want to switch to
HSP/HFP profile, even if the device is already set to HSP/HFP or has already an
input route. Apart from this, we also want to make sure there is no pending
timeout source when we are creating a new one. This should avoid an infinite
loop about switching BT profiles when capture nodes are created and destroyed
quickly.
Indexing into the subject from a node-removed event is slightly quirky.
As a result, we were not properly freeing filter chains tied to
disconnected nodes.
Change how we store the list of loaded filters and ensure they
are properly freed when their parent node is removed from the graph.
Signed-off-by: James Calligeros <jcalligeros99@gmail.com>
If a source filter is used, the BT profile autoswitch does not work because
WirePlumber thinks there is always a client capturing audio from it. This patch
fixes the issue by recursively walking through the source filters that are
linked to the BT loopback source until a stream is found. If a stream is found
at the end of the chain, then it switches to HSP/HFP profile, if the stream is
not found, the profile stays to A2DP.
With this change, it is possible to have a top-level object that does
not start at exactly the beginning of the file, allowing comments and
whitespace to exist.
Also add an empty conf file in the tests to verify that
it does not crash.
- split merge.conf into two files, one being standard JSON,
to ensure we can parse this
- ensure that key-value pairs are correctly overriden when
merging without the override. prefix
- remove context.modules, which is no longer needed there
- fix a typo with a stray ; character
A bit of refactoring was needed in order to avoid breaking as-section
loading, which expects the entire file to be an object or array and doesn't
parse it.
Fixes: #633
In some cases the monitors emit the object_info callback with the same id
and non-NULL info multiple times. For example, when an ACP profile changes from
output:analog-stereo+input:analog-stereo to just output:analog-stereo, it emits
object_info() with NULL info for the input node and object_info() with updated
properties and the same id as before for the output node.
This causes the spa-device here to emit create-object multiple times for the
same object and this breaks the name deduplication logic. To solve this, make
sure we emit object-removed before calling create-object a second time.
Fixes#500
Copy all the node properties into the session item and customize
afterwards, instead of starting with an empty set and cherry-picking
properties... We need to cherry-pick more and more properties
overtime and it's easy to make mistakes
Otherwise it is global and it retains a reference to the `parent`
WpSpaDevice object until this function is called again, which prevents
some camera nodes - in some cases - from being destroyed
Fixes#640
At the moment, the same transition can be completed multiple times
(assuming sufficiently high reference count):
return_error()
finish()
return_error()
The consequence of that is that the transition closure will be
invoked multiple times. This is unexpected.
Add some guards against completing a transition or calling finish()
multiple times.
Fixes#628
Enhance the parsing logic to consider more than one device number. libcamera
devices some times use up multiple V4L2 devices, in this case libcamera should
be given a chance to enumerate the device.
Fixes#623
If the user is reported as active then check that they have at
least one active seat and downgrade the status to online if not.
This ensures that a remote login session won't be interpreted as
the user being active on a local seat.
Make decision on whether to show code location in journal messages based
on whether the WpLogTopic would enable DEBUG level messages.
Add wp_logt_checked API to take WpLogTopic as input to make this
possible, and deprecate wp_log_checked.
Logging messages are usually developed based on the stderr logging,
which shows code location and log topics as context, and especially
higher log level messages can be cryptic or useless without that
context.
This context is not shown in journalctl by default, and so usually is
missing from logs submitted by users. In principle it is available in
journalctl but requires extra work to show it.
Fix by prepending the log topic to the messages, like Pipewire logging
does. Also show code locations if log level is high enough.
This patch makes the check more robust when detecting main filter nodes. This is
because some filters might append '/Internal' to their main node media class.
The direction check has also been improved to work with Video filters.
event:set_data() coerces Lua tables to properties, so that keys and
values become strings. Take into account that profile.index is a string
due to this.
Fixes WP setting the profile, even though it is already the active one.
Currently nothing removes the WpLoopSource from the main context, and we
moreover are creating and attaching unused WpLoopSources when sharing
same pw_context between multiple WpCore.
Fix this by keeping track of the WpLoopSource in the pw_context user
data, only creating them when needed, and finalizing it after the
context is destroyed.
It is intended that the Pipewire loops are "entered" the whole time they
are running, i.e. are going to process further events. Currently we
enter/leave for single iterations, which is not right.
Fix by entering the loop on initial GSource idle callback, and leaving
at finalize.
That the loop has to be left from the same thread it was entered from is
a bit hard to do with GSources, as the finalize callback appears not
guaranteed to be called from any particular thread. I didn't find a
fully general way to do this, so this puts additional constraints on how
WpLoopSource (and WpCore) are to be cleaned up.
