Re: Curious thing about ties

[Alexander Kobel]

On 2016-11-15 09:21, David Kastrup wrote:
> Andrew Bernard <address@hidden> writes:
>
> > Hi Robert,
> >
> > I for one don't think lilypond should be making decisions for you about tie
> > direction based on an idea of sequenced phrases. I know lilypond strives to
> > do as much good layout as it can without requiring tweaks, but this seems a
> > step too far. There is nothing mentioned about this in the NR as far as I
> > know. I would say it is undesirable, and a defect. Others may differ. I
> > would go so far as to attempt to raise this as a bug on the bug list for
> > further evaluation.
>
> Looks random to me, actually.  Like depending on the order of stuff in
> memory.  The kind of stuff which could be different on a different
> computer.  Or even next time you try it.

Just a very vague gut feeling: rounding issues? IIUC, Lilypond 
determines an "uglyness" score [1] for both variants and chooses the 
least unattractive one. If both directions are next to indistinguishable 
(but, e.g., the slightly imprecise horizontal position might be an 
intermediate value in the computation), the outcome will appear random.

Can one see and compare the uglyness scores for the variants considered?

[1]: http://lilypond.org/doc/v2.18/Documentation/essay/automated-engraving


Cheers,
Alexander