[WIP] Repeated accidentals

[battermann]

I only added a test that shows what is to be implemented.

The problem is not trivial at all and I currently don't have a strategy for an elegant solution.

What makes is difficult is that a staff can contain multiple voices, so it is not possible/easy to thread through a state while rendering each voice. Because the current state can affect notes from previous voices that have already been rendered.

Here is an example, which is the desired result:

• The C# from the first voice influences the C from the second voice (the second C has to have a natural sign to cancel the previous sharp)
• In the first voice there is a E natural that also has to be canceled because the second voice has a Eb that comes before it
• In the second bar you can see that the key signature has to be taken into account (just to not forget that)

Currently the result would be this:

Here are the ABC codes:

Expected result:

M: 4/4
%%score (0 1)
V: 0 clef=treble
V: 1 clef=treble
K: G
[V:0]^c2/1 d2/1 z2/1 =e2/1 | z8 |
[V:1]z1/1 =c2/1 d1/1 z1/1 _e3/1 | F2 =F2 F2 ^F2 |

Actual result:

M: 4/4
%%score (0 1)
V: 0 clef=treble
V: 1 clef=treble
K: C
[V:0]^c2/1 =d2/1 z2/1 =e2/1 | z8/1 | 
[V:1]z1/1 =c2/1 =d1/1 z1/1 _e3/1 | ^F2/1 =F2/1 =F2/1 ^F2/1 | 

You can try it in this ABC editor.

It is also the question if this should be done during the ABC rendering. Or does it make more sense to have a processing step before that, where notes are assigned with an additional attribute that indicates wether the accidental is shown or not:

prettifyAccidentals : Notation a -> Notation a

The last resort would be to constrain a staff to one voice???

Or to go for a completely different Notation model.

Depends on #46 so this is set as the target branch ATM to have a nicer diff.

[duncanmalashock]

It seems like one possibility is to use all the voices to generate another datatype that's like a Voice, that just contains a list of changes to the accidentals that are implied at any given point in a staff. We could call those, for example, BarAccidentals, which might include a Duration. At the beginning of a bar, there would be a BarAccidentals containing the notes in the key signature:
C, D, E, F#, G, A, B, Duration Zero
Then after the first note happens (the C# in voice 1), there would be a new BarAccidentals with these notes:
C#, D, E, F#, G, A, B, Duration Quarter
We'd make new BarAccidentals every time an accidental was changed. Once the BarAccidentals data was generated, then it is trivial for the ABC converter to check whether a Note in a Voice had the matching accidental in the BarAccidentals when it's generating output.