[Bf-committers] Quat interpolation and shortest distances problem -- Fixed

[Roland Hess]

Old problem, revisited and fixed...

 From a bf-committers mail of two months ago:
 > While working with Action Baking (translating object-level motion
 > into a bone-level Action), I've run across a small deficiency in the
 > way that Blender handles moving between some quat positions.
 >
 > The problem is that under certain conditions what should be a 10
 > degree rotation along an axis in the course of a few frames turns
 > into a 350 rotation, generating a "flipping" effect. It seems that
 > when smoothly interpolating between two quat positions, Blender on
 > occasion does not choose the shortest route. This problem was noted
 > already in OGRE, and a search turned up this page in their Wiki that
 > references the problem:

According to Ton, this was a key setting problem, not a quat calc  
problem. So, I looked in the key setting function, and added a little  
routine that checks the about-to-be-added quat key value against it's  
predecessor. If it's a flipped value (absolute value close to it's  
neighbor but opposite sign), it's righted. BTW, I tested with in  
conjunction with Action Baking, which makes tons of quat keys, and it  
solved all pose flipping problems!

Simple enough that I hope some intrepid soul commits it!

Bug and patch at:
http://projects.blender.org/tracker/index.php? 
func=detail&aid=4160&group_id=9&atid=125

Roland Hess
harkyman