|
|
Do not output CC#7 events in MIDI on dynamic changes
As volume changes for MIDI are encoded as note velocities, it is not
necessary to add "silent" Audio_dynamic elements to Audio_staffs to
be processed by Midi_walker. (Every "silent" Audio_dynamic event
will only get converted to a Midi_dynamic event outputted as a CC#7
event with fixed volume 100 in MIDI, see Midi_dynamic::to_string.
This behavior is not consistent with the handling of other MIDI
controls on which LilyPond will not enforce any "default" values.)
As Staff_performer::get_audio_staff has the side effect of creating
a new Audio_staff for a voice if it does not yet exist, accessing the
current voice's associated Audio_staff was moved before the handling
of Audio_dynamic elements.
Based on the revision history, not emitting any CC#7 events in MIDI
appears to have been the original intention when encoding dynamic
changes as note velocities (93b7a6ff072d73dcdd41da59cd18da8aa8d8e8cb),
but apparently the initial implementation (of passing "silent"
Audio_dynamic events to Midi_walker, and ignoring them only at the
output stage) caused problems with MIDI timing (#1593), possibly
because the skipped events might have interfered with the logic for
tracking the delta times between the MIDI events that would be
actually outputted. To fix #1593, commit
f114e3c33f9c37c39c7a3fedf66ca5a074785118 conservatively restored the
emission of CC#7 events in MIDI. This commit tries to avoid the
issues with MIDI timing by pruning all Audio_dynamic events already
from the input given to Midi_walker, so that Midi_walker sees no
events, the skipping of which would confuse the measurement of delta
times at the output stage.
This modification (intentionally) changes the MIDI regression test
results by removing all CC#7 events on dynamic changes from the MIDI
output.
|
