A `pw_core` may be shared between multiple streams, device provider
instances, thus when the reference of the given component to the core
is dropped, the event handlers must be unregistered so as to avoid
use-after-free and similar issues.
Fixes#5030
Fixes: 2bc3e0ca10 ("gst: deviceprodiver: Use GstPipeWireCore and some cleanups")
Move the latency print code after where we print the port. That way
we only get the latency when we first print the port.
Avoid -lt from printing latencies for ports without a link.
If in PAUSED state, the node can move from idle to suspended resulting
in format cleared and state is no longer negotiated. To avoid returning
not-negotiated error upon basesrc calling create callback, wait for new
format to be provided and negotiated state is back.
WirePlumber recently added a mechanism to force mono mixdown on audio
outputs, which is a useful feature for accessibility. Let's also expose
that setting via libpulse for existing audio settings UIs to be able to
use.
Pipewire uses a rate of 256/7680 with the integrated camera of Apple
silicon Macbooks. To calculate pw_time.delay correctly in this case it
has to be divided by time->rate.num. Without this division the delay
contribution of the `((latency->min_ns + latency->max_ns) / 2)` term
ends up as 255 which are 8.5 seconds.
pipewiresrc reports the delay as latency in the gstreamer pipeline which
results in rendering a frame every 8.5 seconds.
I suspect the non-normalized rate of 256/7680 is another bug in
pipewire. The rate for an UVC webcam is reported as 1/30. Both
Video4Linux2 devices report a discrete frame interval of 0.033s (30fps).
Fixes#4957
In the current state the GET/SET stream format can handle the commands
response however, yet, it does not take care of checking that:
* A bound input stream cannot have it set, should reply accordingly
* A STREAMING_STREAM output stream cannot have it set, should reply
accordingly.
A flushing invoke is dangerous because the application might have queued
a destroy, which could then be executed right while we do things.
Avoid doing the flushing invoke from the state change function. This was
done because previously we would invoke a process call when we were
working in non-RT mode. Nowadays we run the process function right from
the main thread and we don't need to invoke anymore. This also means
that we can't have pending process calls to flush out when we go to
paused. We do queue other calls, like drained and trigger-done but it
should not cause problems to let those through after the state change.
If this causes problems in the future, we can check the state before
emiting them.
Do a flushing invoke right before freeing the stream. This should be ok
because we removed all signal hooks so that the pending invokes would
not get to the app.
Fixes#5010
The raw mode -a only supported reading raw data from stdin/stdout and
simply ignored the filename. Make it use the filename to determine
where to read from instead.
Support stdin/stdout for sysex mode as well and close the file when we
are done.
Fixes#5012
In the destroy we first remove all listeners and then we clean up the
memory. Move the memory cleanup to a separate function to make it easier
to refcount later.
af3ad7bf9f set the fd and MAPPABLE
flag on buffer data when it was allocated inline in shared mem, even
for MemPtr data. In the client-node however we don't pass along this
fd and so we also should not pass the MAPPABLE flag for MemPtr data.
This fixes an issue in older clients that blindly try to map the fd when
the MAPPABLE flag is set, even when there is no fd (and there is already
a data pointer).
Fixes#5003
