Cross-Staff Beaming

[arshiacont]

This is an excerpt from Debussy's Children's Corner

{[ \staff<1>  
   (* meas. 1 *)  \clef<"f4"> \key<0> \meter<"C", autoBarlines="off", autoMeasuresNum="system"> 
   \staff<2> \stemsUp  \beamBegin:1 \beamBegin:2  f-1/16 c0/16 d0/16 e0/16 \beamEnd:2 \beamBegin:2 
   \staff<1> \stemsDown f0/16 a0/16 c1/16 d1/16 \beamEnd:1 \beamEnd:2 
   \staff<2> \stemsUp \beamBegin:1 \beamBegin:2 e1/16 
   \staff<1> \stemsDown d1/16 c1/16 a0/16 \beamEnd:2 \beamBegin:2 f0/16 
   \staff<2> \stemsUp e0/16 d0/16 c0/16 \beamEnd:1 \beamEnd:2 ]
 , 
[ \staff<2> 
   (* meas. 1 *) \clef<"f4"> \key<0> \meter<"C", autoBarlines="off", autoMeasuresNum="system"> \beamsOff 
   \ten<position="below">( f-1/1)
 ]
  }

Leads to

Whereas it should look like this:

Two issues:

• The obvious beam rendering.. maybe I'm doing sth wrong in the Beam syntax?!
• The duration value of the note in Voice 2 f-1/1

[arshiacont]

@dfober any hints where to start chasing this one in the code?

[dfober]

beaming is computed in GRBeam.cpp, more precisely in GRBeam::tellPosition, the code is really hard to understand (very long methods). I've already add a bit more structure a couple of years ago, but it remains a nightmare.

[arshiacont]

Just for memo: The problem in this issue is related to the initp0 method called in GRBeam::tellPosition which calculates a yoffset, exactly this block:

guidolib/src/engine/graphic/GRBeam.cpp

Line 398 in 6d1fb44

if (setref) {

the yoffset in this context seems to be wrong.

[arshiacont]

Slightly more readable test:

{[ \staff<1> 
  \clef<"f4"> \staff<2> 
   \stemsUp \beamBegin:1 \beamBegin:2 f-1/16 c0/16 d0/16 e0/16 \beamEnd:2 \beamBegin:2 
   \staff<1> \stemsDown f0/16 a0/16 c1/16 d1/16 \beamEnd:1 \beamEnd:2 _/2 ]
 , 
[ \staff<2>  \clef<"f4"> \key<0> empty/1 ]
  }

[dfober]

it looks like the beam is drawn in the wrong context e.g. the coordinates are relative to the first staff and it's drawn with the second one.

[arshiacont]

Thank you @dfober .

It seems like the problematic Beam here is being decoded as smallerBeam in GRBeam. The TellPosition attempts to fix offsets at the end of its function call but they are evaded by this line while treating fSmallerBeam:

guidolib/src/engine/graphic/GRBeam.cpp

Line 1337 in 6d1fb44

return;

What are smallerBeam in this context?!

If I comment that return statement, I get this which is slightly better but still have way to go:

[arshiacont]

@dfober I managed to get the following result by doing two things:

1. Removing the return statement mentioned above so that the offset of the beam is correctly set
2. Forcing the right-most beam [for now manually] as a fSmallerBeam of the longest beam (the first beam is included already but not the second) in GRVoiceManager::organizeBeaming(...) .

Whenever you have some time, we can go over the solution since I'm not sure if I understand the intent of fSmallerBeam and to avoid creating regressions. Ping me over email or else.

Once this is done the long beam's slope/initial points can be fixed more easily given the nested beams.

[arshiacont]

I am half-way done!

A particular challenge would be this excerpt where the nested beams (aka fSmallerBeam) are at the system-level:

{[ \staff<1>  
   (* meas. 1 *)  \clef<"f4"> _/2 \staff<2> \stemsUp \beamBegin:1 
   \beamBegin:2 e1/16 \staff<1> \stemsDown d1/16 c1/16 a0/16 \beamEnd:2 
   \beamBegin:2 f0/16 \staff<2> \stemsUp e0/16 d0/16 c0/16 \beamEnd:2 
\beamEnd:1 ]
 , 
[ \staff<2> \clef<"f4">  _/1 ]}

[dfober]

I've checked the wrong context hypothesis. This is the result when the beam is drawn in the context of the first staff

The different slopes show also that nested beams are treated separately.
My conclusion at this point is that the whole mechanics would have to be revised to handle nested beams properly.

[arshiacont]

Interesting.. Is there a branch with your hypothesis?
On my branch I had a quick look at the slope issue for Nested Beam. in the setBeam() function, the "parent" beam can ideally adopt child positions but my conclusion was that it could easily work on this example but we can come up with more complex compound beaming where it wouldn't... .

[dfober]

No, there is no specific branch. I checked by correcting the y coordinates manually for this specific score (by subtracting the 'y' position of the staff) i.e. GRBeam::OnDraw :

if (st->p[0].y > 380) { ay[0]-=400; ay[1]-=400; ay[2]-=400; ay[3]-=400; };
	// This does the drawing!
	hdc.Polygon(ax, ay, 4);

where 400 is the second staff position().y

[arshiacont]

The changes I made in this PR actually does this: check changes in GRBeam.cpp: removing the return statement will allow the tellPosition to reach the end where we fix offsets.

On top of the above: changes in GRVoiceManager fixes how beams are nested so that the second beam renders as being nested.

Do you remember which issue was being broken in the regression test?

[dfober]

Based on your modification, here is the best result I can get (for now and without breaking the tests)

I find it ambiguous: the second part looks like feathered beams. However I can commit that if it suits you.

[arshiacont]

hmmmm I guess this is better than what we have. Let's commit this. I will then break this into several tickets as I can see what the problems are.

[dfober]

that's done (pushed). I'll go on with this issue.

[arshiacont]

@dfober Here is the unit test that gathers several beaming issues all-together as discussed a few days back:

{[ \staff<1>  
   (* meas. 1 *)  \clef<"f4"> \key<0> 
   \staff<2> \stemsUp  \beamBegin:1 \beamBegin:2  f-1/16 c0/16 d0/16 e0/16 \beamEnd:2 \beamBegin:2 
   \staff<1> \stemsDown f0/16 a0/16 c1/16 d1/16 \beamEnd:1 \beamEnd:2 
   \staff<2> \stemsUp \beamBegin:1 \beamBegin:2 e1/16 
   \staff<1> \stemsDown d1/16 c1/16 a0/16 \beamEnd:2 \beamBegin:2 f0/16 
   \staff<2> \stemsUp e0/16 d0/16 c0/16 \beamEnd:1 \beamEnd:2 ]
 , 
[ \staff<2> 
   (* meas. 1 *) \clef<"f4"> \key<0>
 ]
  }

[dfober]

fixed (could be improved)