Re: New snippet: remove staff if another is alive (for wind divisi)

[David Kastrup]

Thomas Morley <address@hidden> writes:

> 2018-08-07 21:42 GMT+02:00 David Kastrup <address@hidden>:
>> Thomas Morley <address@hidden> writes:
>>
>>> 2018-08-07 20:28 GMT+02:00 David Kastrup <address@hidden>:
>>>> Thomas Morley <address@hidden> writes:
>>>
>>>>>> Ok, let me chime in: I've basically developed some of the low-level
>>>>>> mechanisms for divisi staves.
>>>>>
>>>>> You mean that 'make-dead-when stuff?
>>>>
>>>> Well, that's the internal part of the low end.  But it's usually driven
>>>> by the remove-layer property.  You are not actually using it?
>>>
>>> Nope.
>>> I'll have a closer look at it, though.
>>
>> Maybe it's a failure.  It may well have problems scaling since it
>> creates a single hierarchy.  I know that one tenet or inspiration for
>> that kind of numerical hierarchy was that it was reasonably certain to
>> avoid creating strange loops.
>
>
> I've started trying to understand 'remove-layer by reading the IR,
> doc-strings and playing with /input/regression/divisi-staves.ly
> Afaik no further documentation exists.
>
> As mentioned earlier in this thread I think divisi-Staffs should be
> controled from inside themselves, not from the content for the single
> voices.

What does "from inside themselves" even mean?  You need three different
versions: what should appear in the non-divided staves, what should
appear in the upper staff of a division, what should appear in the lower
staff of a division.  All of these need to be present in source, likely
employing the help of tags.

I think one problem here may be what we consider "divisi staves":
I suspect that your approach considers two voices to be identical
outside of a divisi stave passage.  However, divisi passages rarely
require divisi staves unless the division becomes too complex to be
representable in a single staff.

So for me the principal division is not as much one of unison
vs. divided rather than monophonic vs polyphonic (the former usually
lends itself to single-staff notation reasonably well).  Which means
that the content in "does not need divisi staves" passages still
fundamentally contains three different versions to work with rather than
a single one.

> Thus I changed the regtest like below.
> While it works starting with single Staffs, switching to divisi and
> back later, from a divisiVoice
> It fails the other way round, i.e. single Staffs are never printed.
>
> switchOff = \set Staff.keepAliveInterfaces = #'()
> switchOn = \unset Staff.keepAliveInterfaces
>
> violI=\relative d' {
>   \repeat unfold 44 d4
>   <d g'>2
>   \repeat unfold 58 d4
>   \bar "|."
> }
>
> violII=\relative g {
>   \repeat unfold 44 g4
>   <g d'>2
>   \repeat unfold 58 g4
>   \bar "|."
> }
>
> %% start with single Staffs, switch to divisi and back later
> tstI = {
>   \switchOff
>   \repeat unfold 44 s4
>   \switchOn
>   s2
>   \switchOff
>   \repeat unfold 58 s4
> }
>
> %% start with divisi, switch to single Staffs and back later
> tstII = {
>   \switchOn
>   \repeat unfold 44 s4
>   \switchOff
>   s2
>   \switchOn
>   \repeat unfold 58 s4
> }
>
> divisiVoiceI =
>   \new Voice << \tstI \violII \violI >>
>
> divisiVoiceII =
>   \new Voice << \tstII \violII \violI >>
>
> \score {
>   \new StaffGroup \with { \consists "Keep_alive_together_engraver" }
>   <<
>     \new Staff \with { instrumentName = "Violin I"
>                shortInstrumentName = "V I"
>                \RemoveAllEmptyStaves
>                \override VerticalAxisGroup.remove-layer = 2
>              }
>     \violI
>     \new Staff \with { instrumentName = "Violin II"
>                shortInstrumentName = "V II"
>                \RemoveAllEmptyStaves
>                \override VerticalAxisGroup.remove-layer = 2
>              }
>     \violII
>     \new Staff \with { instrumentName = "Violins"
>                shortInstrumentName = "V I&II"
>                \override VerticalAxisGroup.remove-layer = 1
>                \RemoveAllEmptyStaves
>              }
>     %% control divisi:
>     %\divisiVoiceI
>     \divisiVoiceII

This is a total mess.  The point of remove-layers is that lower layer
numbers override higher layer numbers, so your divisi control has to be
applied to the lower layer number only, making the higher layer number
appear as fallback when the lower layer number has no motivation to do
so.

But you switched around the layer numbering, making the divisi staves
the _higher_ numbers (which only appear when the non-divisi staff
suicides).  And then instead of applying the divisi controls to a
selected group, you just apply it to everything, meaning that when you
use "\switchOff", none of the contexts retain a reason to appear.

>   >>
>   \layout {
>     short-indent = 2\cm
>     indent = 3\cm
>   }
> }
>
> Am I doing something wrong?

Definitely.  First start with the given example.  Then put your divisi
controls _only_ in the divisi staff.  That's the one that is interested
in forcing its appearance.

If you really want to have the same source without tags, you could
override DivisiStaff.keepAliveInterfaces and distribute context aliases
in a manner where the overrides have no effect on the non-divisi staves
without triggering errors.

Separate DivisiStaff and whatever contexts (possibly just as containers
with very few properties and engravers of themselves so that they can
contain various Staff-like contexts) might help in layering the
complexity.

-- 
David Kastrup