Re: Map voices to channels in MIDI output

[Keith OHara]

On Mon, 14 Mar 2011 14:19:46 -0700, Jan Nieuwenhuizen <address@hidden> wrote:

> Keith OHara schreef op ma 14-03-2011 om 13:50 [-0700]:
>
> > Dynamics are newly implemented as note-on-velocity,
> > but the old implementation as channel-volume is still there,
>
> What would the correct fix be?

Well, MIDI note-on-velocity is intended for dynamics.  MIDI channel volume is 
intended for balance, like Lilypond's midiMinimumVolume.
In a development branch, implementing note dynamics purely as note-on-velocity 
would allow MIDI-interested-hackers to use volume for :
1) re-balancing the instruments (because the new implementation messed up the 
equalizer)
2) implementing (de)crescendi on held notes (which sometimes works accidentally 
in 2.12, due I think to notes in other voices creating volume changes on the 
held note)

> We could revert the velocity changes, and have 'voice mode not
> reuse channels...
Very sensible for the stable version.

> > > > Do you want the new MIDI output in 2.14
> > > Yes.
> > Why?
> It's better and newer!

Oh!  That's where we've misunderstood each other.
Lilypond uses an even/odd version numbering system.  2.14 has an even minor 
version number, meaning it is a stable release.  I'm not a programmer, but I've 
worked with them enough to know that better and newer things usually require 
some time before becoming stable.


> So it would be prudent for each new track to always set the
> instrument?  That should be doable.

Prudent.

> In fact, this should do it
>
> diff --git a/lily/staff-performer.cc b/lily/staff-performer.cc
> @@ @@ Staff_performer::new_audio_staff (string voice)
> +  if (!instrument_string_.empty ())
> +    set_instrument (channel_, voice);

Segfaults during midi generation.

Thinking the segfault might be have same cause as what Neil uncovered
      http://lists.gnu.org/archive/html/lilypond-devel/2011-03/msg00375.html
I applied the 'obvious' fix for that problem, which is two lines above your 
suggestion.
This fix allowed make check to succeed, so I pushed 8b9e7a706f .  Please adjust 
if the 'obvious' is not what you intended.

Adding set_instrument() as above still segfaults.  It seems channel_ == 0 at 
this point.