Hi Jean-Paul, yeah unfortunately you will probably run into many things that will not inset, particularly with curved pieces.
The Inset command needs to do a combination of several complex calculations, including a solid offset, then a boolean merge with some coincident surfaces involved, then an additional solid offset that shrinks down the "plug".
There can be problems with any of those steps, but in particular the offset function in the geometry library that I use can only handle pretty simple cases where there is no change in topology in the generated offset.
When you have a face that is completely smooth with its surrounding area, that will require a topology change in the offset. Here I'll try to explain it -
So one part of Inset is that it tries to calculate an offset of the set of faces surrounding your selected one, where the only the selected face is actually offset and the other surrounding faces stay in place, like this:
Then each of those surfaces gets extended and intersected with each other to form connections in the offset as necessary.
But the geometry library offset is set up to expect that for each vertex and edge of the original that there will be a matching vertex and edge in the offset result. That's not always the case though, like when you have a smoothly connecting face, it will try to do something like this:
So note there that when the offset tries to get extended in that case it won't actually intersect with the neighboring face, really some additional faces need to be inserted in there, which means a change in topology which is not currently handled directly by the offsetter.
The Inset command will try to solve this when possible by trying to take the face as a completely separate surface and ignoring its connections to adjacent faces. It's easier for that additional topology to be inserted properly when there were completely unattached edges in those spots. That's the only reason why the other examples on the same surface for the circle and rectangle parts actually worked. Sometimes you may need to manually separate out a smooth surface though.
I think in your case with the ellipse part that it's probably running into the other difficult area of doing the merge with coincident surfaces though, that's also a difficult calculation as well...
Still there may be some potential improvements that I can make, possibly the merging difficulties may be easier to do something about than offset problems.
Currently it will tend to have an easier time with more blocky types of shapes.
- Michael