Parsing MIDNAM is what takes the most time.
"Reset Remote Controls" message is misleading (it means MMC), not
control-surfaces, which a user may have none.
Do not send continuous stream of note-evens with zero velocity
after muting a MIDI track or after a fade to silence completed
(and all-notes-off has been sent).
NB. Automated Faders are left untouched (no panic is sent for -infdB gain),
Note-on events with zero velocity are sent.
This resolves a circular dependency:
libardour calls methods from libardour_cp and vice versa.
Since 9bb2f2bb libardour is also calling active() and that method
needs to be forced to use late binding. -- compare to b9bbea7174
USB stack may not be available on some systems, e.g. unit-test VM.
When libusb_init(0) fails to create default context, further calls
into the libusb API will cause segfaults.
Since 3d15499cda, libevoral enforces Parameter min/max
range. Ardour::ParameterDescriptor sets FadeInAutomation range to 0..2.
Hence all unit-test data needs to be in this range.
ControlProtocols are single instance.
Activating an already active protocol leads to crashes due to
various rasons (e.g. port already registered), re-used singleton
event_loop_name and request-buffers, duplicate free of
AbstractUI request buffers during deactivate,..
When I/O port-counts do not change, plugin re-order happens
in sync in the process-thread. ::configure_io() is only called
to ensure that the current configuration is valid.
In case that the ChanCount does not change, the method must
be realtime-safe and not block.
DiskWriter::reset_write_sources() is not realtime-safe and
implicitly causes a session-save:
Write-sources are destroyed and re-created. This includes
a call to write_source->drop_references(), which triggers
ARDOUR::Session::remove_source(), which saves the session.
Furthermore adding/removing plugins will likewise call
::configure_io().
Previously any processor change on a track lead to
saving the session!
