Our sink and source callbacks are called twice, once when the graph
starts and once when it completes, make sure we don't signal the jack
graph twice or we get corrupted output.
Fixes#3255
Add an option to export the node from the factory.
That way you can locally create a node from a factory and then export it
to the server. This is handy to make object in the server from a config
file using the node-factory.
See !1643
Check for multiple recursive disconnect calls by checking the flags
before calling the pw_proxy methods. This ensure that calling disconnect
from one of the proxy removed callbacks does not trigger another
disconnect later on.
When we have a callback installed but the bufsize/srate can not be
notified yet because the client is inactive, try again in the next
cycle.
Fixes#3297
If we detect Playback/Capture Pitch 1000000, we can adjust those values
to update the feedback endpoint for the host. On the capture side, this
will instruct the host to adjust the rate at which data is being sent.
On the playback side, this will adjust the amount of data the USB gadget
driver sends out in each USB tick.
Let the implementation handle this when needed. The main problem is that
we can set this to NULL before calling the port function and that might
crash things.
If we are not recursing, take the complete property value without
attempting to parse it as json, only do this when recursing.
If the property value looks like a container is is exactly of the right
length, print is as a container, otherwise print is as string.
This makes properties like "[192.0.0.1]:45000" be printed as a string
instead of the array [192.0.0.1]. but still makes "[FL FR]" be an
array.
Fixes#3290
Because the capture triggers the playback stream, the playback stream
might not be actually triggered when the stream is not running.
This can cause a buffer to be queued in the capture side that is never
dequeued from the playback side. If 2 buffers are queued (and 2 buffers
are available on the stream), the capture source has no more buffers and
starts to drop/stutter.
Fix this problem by always dequeueing/queuing all the queued buffers so
that we always use the last one.
See #3276
When one side of the loopback suspends, do a EnumFormat params to force
renegotiation on the other side as well.
Suppose this:
1. sink/filter/pw-play plays at 48000Hz
2. pw-play goes away, sink and filter-sink suspend, filter-playback
idles
3. pw-play starts with 44100Hz
4. sink/filter-sink renegotiate to 44100Hz, filter-playback is still
at 48000Hz and pitch shifted.
We might want to manually suspend the IDLE nodes instead but for now
this is a good workaround.
Fixes#2969
Clear the READ flag when the param is removed.
Don't recursively emit param_changed events or we could end up in an
infite loop when we update params from param_changed.
The rt.mix_lst is really something internal to the tee and fallback
mixer in the ports so make it private.
Use the port_set_io call to add the Buffer io area to the mix_list
tee and fallback mixer on the port, like we do for remote-node. We can
then remove the custom code to do this in remote-node and impl-link.
Remove an unused field (clock) in the port struct.
Remove the unused io_set field in impl-link, it is always in sync with
the activated field.
Clear the callbacks from the processing thread to avoid races.
Check the callbacks in the processing thread before calling them,
we just need to check if there are functions, we checked the method when
installing the function.
Fixes#3251
@DEFAULT_MONITOR@ finds the default sink but returns is_monitor true.
Always lookup the device based on the name and index. This transform
the special device names like @DEFAULT_XXX@ to the default device.
Always use the found device name as the target. Make sure to set the
CAPTURE_SINK property when dealing with a monitor.
Fixes#3284
On musl, `pthread_t` is a pointer type, so printing it
as a long generates warnings. So cast it to `void *`
and print it with "%p" since it is actually a pointer
even on glibc (nptl) (at least at the time of writing...).
Make NTP timestamps based on CLOCK_REALTIME.
Handle socket errors.
Some devices want at least 1 second of latency between RTP and NTP
timestamps or they stay silent. A a raop.latency.ms property for this
purpose that defaults to 1 second.
It is said that all devices seem to add 250ms of extra playback delay,
so include that into the delay reporting.
Fixes#3247
