Maybe I'm just misunderstanding this slide, but can't we just set the normal in the correct orientation in this case instead of refusing to collapse the edge?
@jfondrie It's actually not that simple, because we are not defining normals explicitly. In order to change the normal of a face, we need to move the vertex (computed by minimizing the quadric) to some other location. But what is the "best" vertex in scenarios like this one? It's not obvious. So the workaround proposed here is to simply avoid collapsing these edges altogether.
Does that mean when we should compute the dot of all the edge pairs connecting to the new vertex? Also, normally when we compute this dot, we've done collapsing already. Does that mean that we have to undo the collapse? Or is it like a pre-checking?