Only basic midi ports use the _trace_parser and they enter
here via cycle_start which will pass the _trace_parser to
this function. No need to double up calls.
BindingProxy owns a PopUp prompter with 30 sec self-destruct
timeout. However ~BindingProxy explicitly destroys the PopUp.
Later the timeout can lead to a double-free.
This fixes a crash when deleting a plugin within 30 sec
after using MIDI learn.
MidiControlUI drops the reference to a given controllable in
its own thread. This can happen after the plugin is already
destroyed (even though the PBD::Controllable still exists).
This uses the existing optional runtime ffmpeg dependency to add support
for m4a files (and in theory whatever other file formats ffmpeg
supports) to the import audio dialog.
The same functionality is mostly already available via Session -> Open
Video, with the "import audio only" selection (even though m4a isn't
currently included as one of the video formats, it still works). Having
this in the import audio dialog however seems much more user friendly.
Harvid daemonizes and does not write anything
to stdout/err. as opposed to select(), poll() on
macOS does not return when the child process
terminates or is killed.
However poll() on an invalid FD does throw an
error and POLLNVAL is set.
There is no need to connect signals twice, can connect them directly to the
helper UI loop and skip the surface loop. Then let the server decide if it is
necessary to call lws_cancel_service() or not.
Also rename WebsocketsServer::should_request_write() to read_blocks_event_loop()
it makes more sense for the caller now on-demand write logic is completely
implemented by the server class.
In between timespans Ardour disables freewheeling and
only later resumes freewheel. It can happen that
Session::process can summon the butler while start_audio_export()
calls Track::seek.
Just waiting for an earlier Butler::wait_until_finished() does
not prevent the butler from running again later.
This can lead to concurrent calls of DR::refill_audio from both
the butler and start_audio_export(), leading to garbled buffers.
ExportHandler::start_timespan() is also used to stop export
when there are no more timespans to be processed.
This is done because freewheeling has to be stopped from
outside the process cycle.
This is an update to the surface ArdourFeedback class that is needed to support
the new event loop integration method.
The various session event callbacks cannot be queued in the surface event loop
because that would create a delay between the time such events are fired and
the time for writing to clients arrive, due to lws_service() blocking while
it waits to read. To solve this issue a helper AbstractUI is created for
catching events as soon as possible and issuing a call to lws_cancel_service().
See WebsocketsServer::glib_idle_callback()
Some distro repositories offer versions of libwebsockets that have not been
compiled with LWS_WITH_GLIB or LWS_WITH_EXTERNAL_POLL enabled. For such cases
a different event loop integration method is needed.
True for Ubuntu 20.04 as of Jun '21
Port_connect must be called with correct order (src, dst),
and may only be called after the client is active.
jack2 reported errors:
> Destination port in attempted (dis)connection of ardour:physical_input_monitor_enable and system:capture_1 is not an input port
> Cannot connect ports owned by inactive clients: "ardour" is not active
Bounce (Session::write_one_track) uses bounce_chunk_size = 8192
to process data. This requires gain_automation_buffer, and
scratch buffers that can hold 8k samples.
Previously the buffers were not allocated when bounding MIDI
(no synth), leading to a crash when automation was used.
Previously there was only a PostDisconnect() signal.
When a port connection is made directly e.g. recorder_ui.cc
r->input()->audio(..)->connect(...), or when connecting
using an external jack control app, Ardour::IO is not informed.
no IOChange::ConnectionsChanged signal is emitted and
Track::set_align_choice_from_io() is not called. This leads to
incorrec capture alignment.
To minimize signal emission the static global PostDisconnect
is changed into a per port ConnectedOrDisconnected signal.
IOs subscribes to signals from owned ports only.
Rely on previously collected I/O port-list, skip expensive call
to get_physical_inputs() and rely on RCU port list alone.
Also skip port_is_mine() test in optimized builds.
* You can now show a selector for each track, rather than one global dialog
* Removed the 'cancel' button: it was too easy to unintentionally revert your selection
* Correctly sort and name new playlists when they are added or renamed from the P menu
* Fix problem where list contents disappeared when changing desktops (on_unmap_event)
* avoid memory allocation (remove std:set<>&)
* skip nested loop using port-names as IDs
However this adds a bit of extra cost in case input ports are
monitored: get_port_by_name() and get_buffer()
glib atomic operations include a complete act as a full
compiler and (where appropriate) hardware memory barrier.
std::atomic<> allows dedicated acquire and release barrier
semantics which are not available with glib.
* Add an option to use a playlist without stealing its ownership. (libardour)
* Tracks emit PlaylistAdded when a new playlist is created either by copying or by creating a new empty pl
* no version-check should be necessary because when opening an old session,
playlists can be looked-up by discrete names and will be tracked by ID thereafter
