issue 5312: Key cancellation glyph position inconsistent

issue 5312: Key cancellation glyph position inconsistent

file lily/key-signature-interface.cc

Using two intervals representing the "vertical skylines" of two
adjacent natural glyphs. This makes a better model of the actual
glyph shape than the original single point within an inverval.
Now we can distiguish between all three cases:
(1) no overlap -> no additional kerning needed
(2) just touching (overlap, but intesection length 0) -> kerning 0.15
(3) overlapping -> kerning 0.3

Case (2) is the new one that couldn't be handled before:
Just touching at the corners made the glyphs cling together.

Basically same  interval technique as before, but I
renamed the Slices from 
  pos and overlap_pos 
to
  ht_right and last_ht_left.
Using quarter stave-spaces as to get a better (integer) resolution.

[Be-3, 2018-04-24 17:25:06]

Please review...

Thanks,
Torsten

PS: no new regression test needed, there are loads of key cancellations in the
existing regression tests.

[Carl, 2018-04-24 17:59:31]

LGTM.  I am just a *little* bit concerned about having the dimensions of the
Emmentaler natural glyph hardcoded in the source, but we already have magic
numbers reflecting the characteristics of the Emmentaler glyphs.

Maybe it would be good to put a FIXME in recognizing this fact.  Or maybe we
just go as-is.

[Be-3, 2018-04-24 18:43:45]

On 2018/04/24 17:59:31, Carl wrote:
> LGTM.  I am just a *little* bit concerned about having the dimensions of the
> Emmentaler natural glyph hardcoded in the source, but we already have magic
> numbers reflecting the characteristics of the Emmentaler glyphs.
> 
> Maybe it would be good to put a FIXME in recognizing this fact.  Or maybe we
> just go as-is.

Hi Carl,

The intervals are just *approximating* the outlines of a run-of-the mill
natural glyph. I even played around with the concept using squared paper.
This approach more or less relies on the fact that the square/parallelogram
part of a natural glyph will be one stave-space high independent of the font
used, just like a notehead in *any* font will be one stave-space high.

Therefore, it will also work for Gonville, Bravura, all of Abraham's fonts...
Funny enough, it works for the LilyJAZZ font, too, because even in this pseudo 
handwritten font the square/parallelogram part has the same (standard) height.

All the best,
Torsten

[Carl, 2018-04-24 21:49:37]

On 2018/04/24 18:43:45, Be-3 wrote:

> The intervals are just *approximating* the outlines of a run-of-the mill
> natural glyph. I even played around with the concept using squared paper.
> This approach more or less relies on the fact that the square/parallelogram
> part of a natural glyph will be one stave-space high independent of the font
> used, just like a notehead in *any* font will be one stave-space high.

So it's a reasonable approximation that the box part of the natural will extend 
about 3/4 staff spaces above the staff position of the natural, and the
descender 
will  descend about 1.5 staff spaces below.

Seems like a reasonable estimate to me, even if it's not exact for a particular
font.
You can't get too far away from that and still fit in a staff.

Thanks,

Carl

[Be-3, 2018-04-24 22:08:39]

On 2018/04/24 21:49:37, Carl wrote:
> On 2018/04/24 18:43:45, Be-3 wrote:
> 
> > The intervals are just *approximating* the outlines of a run-of-the mill
> > natural glyph. I even played around with the concept using squared paper.
> > This approach more or less relies on the fact that the square/parallelogram
> > part of a natural glyph will be one stave-space high independent of the font
> > used, just like a notehead in *any* font will be one stave-space high.
> 
> So it's a reasonable approximation that the box part of the natural will
extend 
> about 3/4 staff spaces above the staff position of the natural, and the
> descender 
> will  descend about 1.5 staff spaces below.
> 
> Seems like a reasonable estimate to me, even if it's not exact for a
particular
> font.
> You can't get too far away from that and still fit in a staff.
> 
> Thanks,
> 
> Carl

Agreed, Carl,

And, IMHO, there is no need for exaggerated exactness: as soon as two 
corners get too close (no matter how close exactly), we need a wee bit
of extra padding.
The main reason for the 3/4 staff spaces, to be honest, was that the two 
intervals will be just touching (with intersection length 0) for the 
corner-to-corner constellation.

All the best,
Torsten