[postgis-devel] ST_Perimeter and ST_Perimeter3D what should Polyhedral Surface return?

[Nicklas Avén]

> After our discussion, i really think that having only LWGEOM struct
> for connected surfaces (PolyhedralSurface, TIN) is not enough.
> 
> We will spend a lot of time/energy in each spatial function to compute
> again the topology beneath.
> 
> There's a need to have a point/edge/face model.
> Need to explore this way...
> 
> --
> Olivier
> 

When thinking about distance-calculations I have had two different
thoughts that I cannot get together.

1)	To have some sort of topology so every connected point is only stored
once, like the points in the linestring. But for three dimensions the
order of the points is not enough to keep them in order. 
2)	To define every plane and then describe the polygon on the plane. I
mean the polygon described like a 2D polygon but then tthe plane
definition tells how the polygon is twisted in space. Then we would
avoid the issue about the inaccurate plane. This approach is like when
drawing a 3Dmodel in som software you draw it on the 2D screen and then
twist the object to draw on another plane.

But the two ways of thinking about it I cannot get toghether. If one
vertex that is a member of two different polygons in two different
planes shall be stored only once, then I cannot see how to get the good
part out of the second approach. 

What we want is both to keep the topology in the connected points AND
keep the planes absolutely plane. Is that possible to build into the
model? That would be very nice and worth some thinking I think.


To keep the planes plane is not only a matter of the calculations I
guess but also about how to build the 3D objects to get them valid in a
good way. It would be nice if valid objects only could be created using
special client software.

/Nicklas