lilypond-auto
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Lilypond-auto] Issue 1713 in lilypond: Whole note/dotted note colli


From: lilypond
Subject: Re: [Lilypond-auto] Issue 1713 in lilypond: Whole note/dotted note collision in polyphony
Date: Tue, 15 May 2012 17:16:33 +0000


Comment #30 on issue 1713 by address@hidden: Whole note/dotted note collision in polyphony
http://code.google.com/p/lilypond/issues/detail?id=1713

@Pavel (comment 22):
"The patch removes several constants [...] but I don't think it's a good approach for Lilypond."

Generally i agree, but not in this specific case. The original code is very hard-wired and quite broken. It produces more bad (completely bad, not just imperfect) results than good ones.

"A generic algorithm makes it harder to tweak one distance without tweaking other distances."

but a generic algorithm makes it easier to change many things in a coordinated, coherent way.

"I believe there should be no slight changes."

If the original code produced at least acceptable results, i would agree. But it didn't. Even if Keith's patch makes 5% of cases slightly worse, it makes half of cases significantly better. In my opinion we should accept it, and then improve it.
compare
http://lilypond-stuff.1065243.n5.nabble.com/file/n5705591/close-collisions_current.pdf
with
http://lilypond-stuff.1065243.n5.nabble.com/file/n5705591/close-collisions_new.pdf

"We need a big test, a matrix with all combinations of notes that can collide."

I thing the above one is good. The source is in a comment in Rietveld http://codereview.appspot.com/6189048/diff/9003/input/regression/collision-whole.ly#newcode16

"That should cover all styles, all standard shapes of noteheads, all durations and all pitch differences."

Are you sure? A matrix like the above one (or bigger) for 15 styles of noteheas and about 15 special shapes used in various shapeNotes? That's about 1000 cases (or more).

I'm not opposed to having an algorithm that would handle all collisions of notes with any shape. But that surely cannot be done with "just-right human tweaked numbers"; we would have to teach Lily recognizing the actual shape of the noteheads. I think it is possible to do this: Mike's work on skylines is very promising in this regard. If you haven't followed the discussion about it, see issue 2148, http://codereview.appspot.com/5626052/ and compare outlines in these files:
http://lilypond-stuff.1065243.n5.nabble.com/file/n5705592/Eja-Mater_master.pdf
http://lilypond-stuff.1065243.n5.nabble.com/file/n5705592/Eja-Mater_skylines.pdf




reply via email to

[Prev in Thread] Current Thread [Next in Thread]