From Andrei Gheata:

I took a look to the way a rotation is
interpreted from Euler angles and I found that the computation of matrix
elements is not as it supposed to be, resulting in a confusing violation
of the documented convention.
  This is why I fixed it (putting also in the computation of the inverse
matrix)
  I touched also TGeoManager.cxx, fixing a warning (variable not used)


git-svn-id: http://root.cern.ch/svn/root/trunk@7988 27541ba8-7e3a-0410-8455-c3a389f83636