This removes the constant checking for state changes with
5ms of sleep in between for the state thread.
We need a new thread to wakeup the state thread reliably
without blocking in the audio thread. For a more detailed
and theoretical explanation of the problem and solution (specifically
written for this commit), see:
https://nyorain.github.io/lock-free-wakeup.html
Now, will only do this time-based sleeping when actively waiting
for a state change. We can't implement that with a blocking call to
AAudioStream_waitForStateChange since that can't be woken up
and we furthermore might have to wait for multiple streams at once.
This also fixes some issues and race conditions with stream destruction
and adds some more documentation.
* ossaudio (FreeBSD currently) backend
The ossaudio backend was rewritten from the sunaudio
backend.
* Fix leaking of mixer_fd in oss_enumerate_devices
* Both input/output can be detected at the same DSP
* Update on device detection during enumeration
* Remove spurious calls to SNDCTL_CARDINFO
* Fix string allocation detection in oss_enumerate_devices
* On FreeBSD, try to open and probe the device's capability in oss_enumerate_devices
* Rewrite oss_io_routine() to implement correctly for all stream directions
* Add preferred device detection for FreeBSD
* Emits stable devid from oss_enumerate_devices on FreeBSD
* Fix SNDCTL_DSP_GETI/OSPACE calls in oss_stream_init
* Use /dev/sndstat instead on FreeBSD for oss_enumerate_devices
* Unify both play and record nfr and take the minimum value between them
* Fix allocating both input/output buffers to input/output-only streams
* Fix clipping issue
* Fix misuse of input_device and output_device in oss_stream_init
* Fix builds on Illumos
* Code refactoring on oss_enumerate_devices
* Improve oss_io_routine for the case when the stream is in both direction
* Use fragsize instead of total buffer size for number of frames
* Probe OSS audio availability in cubeb_init when backend is not specified
* Fix scan-build10 report on src/cubeb_oss.c:285
* Add __DragonFly__ pp macro testing along-side __FreeBSD__
* Move oss_init in default_init[] table right below alsa to respect POLA
* Fix use-after-free of s->mutex in oss_stream_destroy
* Fix inconsistent indentation
* Remove blocks_written from cubeb_stream
* Add LL integer suffix to make 0x80000000 more clear
* Add parsing of /dev/sndstat for hw.snd.verbose > 0 in FreeBSD
* Do device setup in the order of channels, format and speed, according to developer/callorder.html of OSS v4.x API reference
* Add proper latency_frame support for oss_stream_init
* Mark close brackets and close braces NUL in oss_sndstat_line_parse
* Search close brackets and close braces from the end of line in oss_sndstat_line_parse
* Use 32 frames for each fragments in oss_calc_frag_params
* Drop unnecessary #include <sys/sysctl.h>
* Add support for cubeb_channel_layout
* Compilation and scan-build fixes:
* Fix warnings in oss_chn_from_cubeb
* Include cubeb_mixer.h to have cubeb_channel_layout_nb_channels
* Fix potential resource leakage in input_stream_layout->layout does not
match input_stream_params->channels. (The same for
output_stream_layout)
* Classify cubeb_stream_params::layout and cubeb_stream_params::channels
mismatch as CUBEB_ERROR_INVALID_PARAMETER
* AUDIODEVICE can now override /dev/dsp
AUDIODEVICE was chosen to match the sndio backend.
* Change environment variable AUDIODEVICE to AUDIO_DEVICE in oss backend
* Change the format of cubeb_device_info's friendly_name
This avoids name collision when setting media.cubeb.output_device in
about:config of Firefox. However, I think it makes more sense on
Firefox's side to use device_id instead.
* Fix warning of a missing %d in LOG() in oss_io_routine
* Do not enable OSS compilation when SOUND_VERSION < 0x040000
* Add mutex to serialize access to cubeb_strings in cubeb_oss's cubeb_context
* Change the calculation of fragsize/nfr to match stream latency, with 2 fragments available. By HPS
* Now reuse the same thread for audiostream
* Redo the duplex logic
* Restructure the use of buffers
* Fix problems dealing with draining
* Revert "Fix problems dealing with draining"
This reverts commit 30301ba101.
* Revert "Restructure the use of buffers"
This reverts commit 9823ca889c.
* Revert "Revert "Restructure the use of buffers""
This reverts commit 10dc869b85.
* Revert "Revert "Fix problems dealing with draining""
This reverts commit 9d19b10944.
* Make sure there are at least s->nfr available before calling data_cb
* Kill a signed/unsigned comparison
* Add The FreeBSD Foundation copyright line for recent commits
* Fix race condition when restarting a stream by HPS
Originally, when stopping a stream and immediately starting the stream
again, the start call might be lost.
* Do not signal the doorbell_cv again when a stream is already stopped
This is redundant.
* Revert "Do not signal the doorbell_cv again when a stream is already stopped"
This reverts commit fbdf753fc7.
* Make state changes look more similar to pulse backend
This also allows drained/short-input stream to be resumed without
stopping it first.
* Remove a spurious s->running = false
The number of xrun fragments to skip is added once to ctx->jack_xruns but
then subtracted multiple times (once for each stream) from the same variable.
This causes the unsigned int to "underflow" and wrap around.
- Use unsigned int in all calculations instead of mixing un/signed.
- Remove erroneous and useless subtraction logic
- Replace for loop by simple multiplication
- Remove unneeded casts to float
Paul Adenot
nyorain
Vitor Kiguchi
Richard Dodd (dodj)
khng300
Admiral H. Curtiss
Brendan Early
Guillaume Besson
Zoë Sparks
Angelo Haller