Re: Issue 37 - new work

[Mike Solomon]

On Jan 28, 2011, at 1:15 PM, Han-Wen Nienhuys wrote:

> pressed send too soon.
> 
> On Fri, Jan 28, 2011 at 4:14 PM, Han-Wen Nienhuys <address@hidden> wrote:
>> On Fri, Jan 28, 2011 at 11:22 AM, Mike Solomon <address@hidden> wrote:
>>> 
>>> I cooked up this musical example that shows both responses to upward and 
>>> downward pressure to give you an idea of where I'm coming from.
>>> 
>>> Is there a way to get this type of collision avoidance w/o a 2nd quanting 
>>> pass?
>> 
>> You are now dramatically extending the problem scope: you are looking
>> for a configuration that optimizes configurations of all beams at the
>> same time.  For example, it is conceivable that individually, the beam
>> for A below staff would be better served when the beam goes up the
>> line of D''.  However, that would leave no room for the A' beam, since
>> it would be on top of the A'' beam.
>> 
>> A solution that would work for this should optimize all of the beams
>> at the same time.  It might be feasible to code, but probably only if
>> you restrict all the beams to be
> 
> horizontal.  The tie formatting is similar: it also tries to find a
> configuration where *all* ties between a chord are more or less OK.
> 

I've made an updated version where the collision penalty controls the extent to 
which the beam approaches the original quants.  In the png below, the penalty 
is set normal first, then insanely high.
"Normal" builds the regtest to look as pretty as possible but fails in tight 
situations like the first example in the png.  If you set it too high, it leads 
to the non-conventional quanting in the second example.  That said, I think 
that most users would opt for ugly quanting over smooshed beams and thus go w/ 
the second example.

Interestingly, if you comment out the slope damping function, it leads to even 
better results for the inner two beams, leading me to believe that there should 
be a switch that allows one to bypass this function.

Perhaps that should be a separate patch, though.  Thoughts?

http://codereview.appspot.com/4022045

hanwen7.ly
Description: Binary data

0004-Allows-for-variable-collision-detection-in-quanting.patch
Description: Binary data

0003-Tweaks-regtest-and-adds-avoid-collisions-property.patch
Description: Binary data

0002-Triggers-second-quant-pass-for-collisions.patch
Description: Binary data

0001-Implements-a-more-robust-solution-to-issue-37.patch
Description: Binary data