"While 'atomic' has a volatile qualifier, this is a historical
artifact and the pointer passed to it should not be volatile."
Furthermore "It is very important that all accesses to a
particular integer or pointer be performed using only this API"
(from https://developer.gnome.org/glib/2.68/glib-Atomic-Operations.html)
Hence initialization of atomic variables is changed to also use
this API, instead of directly initializing the value.
This also fixes a few cases where atomic variables were
accessed directly.
see also libs/pbd/pbd/g_atomic_compat.h
PortAudio uses what it calls 'default suggested latencies' but in callback streaming mode, they can result in wildly inaccurate buffer sizing (e.g. the user requests a buffer size of 128 but PortAudio actually instructs ASIO to use a much bigger size).
What we do now is to improve PortAudio's suggested latency calculation by basing it on the actual buffer size requested by the user.
The backend holds `_port_callback_mutex` while disconnecting ports.
In some cases disconnecting a port can drop the last reference
resulting in a port-deletion from the connection handler.
This in turn will eventually aquire the `_port_callback_mutex`
and deadlock.
This is now circumvented by using atomic operations instead of
taking a lock to set the `_port_change_flag`.
The flag is also used to trigger a latency update in some cases,
atomic is preferable to taking a lock to set this flag.
--
Full bt: https://paste.debian.net/1184056/
Short:
#1 in pthread_mutex_lock ()
#2 in ARDOUR::PortEngineSharedImpl::port_connect_add_remove_callback()
#3 in ARDOUR::BackendPort::~BackendPort()
#4 in ARDOUR::DummyPort::~DummyPort()
#6 in ARDOUR::DummyAudioPort::~DummyAudioPort()
#7 in boost::checked_delete<ARDOUR::BackendPort>(ARDOUR::BackendPort*)
#12 in boost::shared_ptr<ARDOUR::ProtoPort>::reset()
#13 in ARDOUR::Port::drop()
#14 in ARDOUR::Port::~Port()
#15 in ARDOUR::AudioPort::~AudioPort()
#17 in ARDOUR::AudioEngine::add_pending_port_deletion(ARDOUR::Port*)
#20 in boost::detail::sp_counted_base::release()
#37 in ARDOUR::PortManager::connect_callback() at libs/ardour/port_manager.cc:788
#38 in ARDOUR::DummyAudioBackend::main_process_thread() at libs/backends/dummy/dummy_audiobackend.cc:1018
The warning "samples per period does not match." never triggered.
Previously not being able to set the requested buffersize was a
fatal error.
This adds support for soundcards that only support msec.
e.g. recent HDA Intel via SOF (Sound Open Firmware)
This allow to restore original engine port-names as set
by the backend. ALSA MIDI, CoreAudio, CoreMIDI and PortAudio
drivers can provide human readable physical port names for
some devices.
When exporting long sessions with freewheeling, pulseaudio
may meanwhile suspend the corked audio device. The "FAIL_ON_SUSPEND"
option then prevents ardour to uncork it after export, and the
audio-backend is halted.
This potentially breaks various assumptions (e.g. no resampling,
fixed buffersize) when the stream is moved to a different device.
Then again it's pulseaudio, which is unsuitable for pro-audio to
begin with.
This fixes an issue with some soundcards e.g. "AxeFx III".
Device configuration fails unless set_hwpar() is performed
for the capture device before configuring the playack
device (half duplex is fine, too).
This is mainly for RME RayDAT that has a fixed buffersize of 16k:
dev_name : hw:HDSPMxc2f6c5,0
channels : 36
min_rate : 32000
max_rate : 192000
min_bufz : 16384
max_bufz : 16384
min_nper : 4
max_nper : 512
However nperiod configuration determines the effective latency
regardless.
This is similar to https://github.com/jackaudio/jack1/blob/master/drivers/alsa/alsa_driver.c#L476-L486
This adds a basic support to use multiple sound-cards, currently
limited to two devices: In/Out with shared settings.
Advanced setups still have to resort to using the ARDOUR_ALSA_EXT
environment variable
This is intended to prevent crashes when unregister_port() modifies the contents of these
two members at the same time that something else is iterating over them.
This can help when running with very low latency and the
initial process callback is [indirectly] expensive.
E.g. load a heavy session the a RPi4, initial setup can pull
in a lot of data, which blocks the bus.
In particular with the ALSA backend this can lead to poll timeout
which effectively stops the backend.
When recover() successfully re-initializes the device,
processing can continue just like after an x-run.
This can happen during initial session load of "expensive"
sessions (in particular on slow systems, e.g. Raspberry Pi)
usually with synths. Worker thread pulls in many external
files in the background which blocks the bus for a long time.
resulting in a poll-timeout.
Previously this was inherited via PBD.
On MacOS/X, this adds
"-undefined dynamic_lookup -flat_namespace"
and various "-framework .." options to linkflags
Without this flag, .dylibs fail to link usually because
of missing `-lintl` (Undefined symbols: "_libintl_dgettext")
On other systems this is a NO-OP:
CFLAGS_OSX, CXXFLAGS_OSX and LINKFLAGS_OSX
are only set on the darwin platform.
This produces synchronous events on Audio and MIDI ports.
One rvent per second, exactly at every second since engine-start.
MIDI: C-4 Note-on/off (1 sample long)
Audio: +1/-1 transition:
+1 in sync with Note-on,
-1 in sync with Note-off
gcc can recognize various regexps in comments. Since C++17 provides
[[fallthrough]], using /* fallthrough */ consistently seems
appropriate until we switch to C++17.
see also https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html
Theoretically this could be lock-free by using a queue of device
ports to be added/remove in sync in the process-callback, but
realistically adding/removing devices doesn't have to be rt-safe.
