Hi@whkarto,
One of the major differences between the 2017 and 2015 output that I noticed,is the fact that each building in the 2017 version is a MultiSurface that contains a lot more Mesh parts than the 2015 one.I guess your Building.ffs file contains the original data,right?In that case,the 2015 output seems to be correct (as you expected) and the 2017 one converts the CompositeSurface parts of the MultiSurfaces into more Mesh parts than it originally contained.My guess is that the Collada writer (which the KML writer calls) is to blame here and something goes wrong with the triangulation.
As a workaround,you could try to take care of the triangulation yourself (so the Collada writer doesn't screw up) and deaggregate each building one level so you end up with all CompositeSurface geometries,pull them through aTriangulator(which outputs Meshes) and then reassemble them into MultiSurfaces (e.g.grouped by UUID) using anAggregator.Then write out to KML again.I'm using 2017 and for me,this worked!
Nevertheless,this feels like a bug,so I would report this to@support(e.g.@DaveAt亚搏在线Safe).