On 2011/12/21 16:13:13, dak wrote:
http://codereview.appspot.com/5504055/diff/1/lily/slur.cc File lily/slur.cc (right):
http://codereview.appspot.com/5504055/diff/1/lily/slur.cc#newcode388 lily/slur.cc:388: vsize N_BOXES = 100; On 2011/12/21 15:59:52, Neil Puttock wrote: > Don't you think this is a bit extravagant? At least, this shouldn't
be
> hard-coded. A coarser integral should suffice in most cases I'd
think.
Actually, the convex hull of control points and end points is strictly
outside
of the bezier curve itself and should almost always be a reasonably
good fit. Yes, but our skylines currently consist of boxes that have horizontal tops. So we need to approximate a non-horizontal line (e.g. from control point 0 to control point 1) with a series of horizontal lines. The new patch set sets a modifiable number of 20 for the number of boxes. I'd think that something like 6 ought to work quite well. Mike, have you done any work to examine the effect of the value of n_boxes on a) computation time, and b) quality of engraving fit? If there is no effect on computation time, then we might as well err on the side of too many. If there is a significant effect on computation time, we ought to be careful to set the number as small as possible. Also, in keeping with the desire to use full-word variable names, perhaps box_count would be better than n_boxes. http://codereview.appspot.com/5504055/ _______________________________________________ lilypond-devel mailing list lilypond-devel@gnu.org https://lists.gnu.org/mailman/listinfo/lilypond-devel
