A/V sync is now perfect regardless of the output device (bluetooth,
wired, speaker), and client side interpolation provides high accuracy
regardless of the buffer-size that the users decides on.
Also clear out the memory used by the async logger when it's shut down.
This relies on the fact that enabling and disabling logging is performed
on the same thread, but I think this is a reasonnable thing to do. We
could assert it though.
This covers some callbacks, but not all of them: the APIs that don't
require explicitely creating the thread will see their thread registered
separately.
It's useful to be able to enable logging dynamically while the program
is running, and this can now be done on any thread.
Various threads are logging (directly or asynchronously via the ring
buffer), it's better to have those atomic.
Both values are always checked before logging, and both must be non-null
to log.
Not the prettiest but it does the job. Thanks to @ashleyz we don't really care about checking the log level here, it's checked ahead of time, which is nice because we don't want to carry it in the async log payload.
Potentially significantly shrinks the list of devices to search when
matching BT output device with specified BT input device. On my local
machine, this shrinks the list from 25 to 7, and reduces the initial
enumerate time from ~67ms to ~32ms (and a subsequent call from ~14ms to
~12ms).
Devices in this state are not visible in the Sound Control Panel or
other typical places in the OS audio UI, so there's little point in
including them in cubeb's device list.
On my test machine, this reduces the number of devices enumerated from
118 to 25 - reducing the cost of an initial cubeb_enumerate_devices from
~215ms to ~67ms (and a subsequent call from ~22ms to ~14ms).
- Rather than "leak" an atomic bool to the stuck render thread, we
instead "leak" the entire stream and leave the render thread to clean
up.
- render thread now checks `handle_emergency_bailout()`, which frees the
stream and exits the render thread in bailout condititions.
- Removed `thread_ready_event`, since it's no longer needed.
* Don't change `output_device_id` once it's set
* Fallback to use default device only when not specifying a device
We should only fallback to use default device if the user doesn't
ask to use a particular device.
* Follow default device changes only when using default device explicitly
We should get the default device change notifications only when users
ask to follow the default devices.
* Don't reinit stream when user-selected device is unplugged
If the user specifies a device explicitly, don't switch device for
the stream when the selected device is gone. We should fire an error
callback instead.
* Rename `picked_output_device_id` to `selected_output_device_id`
* Rename fallbackable to allow_fallback
* Correct the comments
* Fix double free problem
unique_ptr can not be initialized from another unique_ptr's raw pointer,
otherwise we will free the pointee twice!
* Add an util to copy wide-char string
* Prefer move
This might be overkilled, but it expresses the intention clearly
Paul Adenot
Matthew Gregan
Ty Lamontagne
Alexandre Bouvier
ivan-matveev
az
japm48
-
aeiouaeiouaeiouaeiouaeiouaeiou
Kevin Yin
Chun-Min Chang