Basically if you have overlapping triangles at a pixel (technically a 2x2pixel area) the renderer has to consider all of them, which means it has to process that bit of the screen repeatedly It's not worth worrying about if the number of pixels involved is small or if preventing it means extra work, sacrificing modularity…
I'd generally advise against that sort of structure unless you really are trying to minimise tricount (eg for a low Lod mesh) the overlapping geometry can have a bigger effect on frame time than you might think - especially if it covers a lot of screen pixels. At a distance the above example would be fine, just not up close
it's up to you what's good and bad geometry depending what you' re working onif as few tri as you can is what you aim for I see this you could do ? and disconnect it from the above piece. no one will see :) maybe push the verts a little bit in so you dont see seams from far or cause shadow glitch its an example it applies…
Can you elaborate on this more? When I model, especially windows and doors, I find it easier to break things up sort the way that post was suggesting. I try to stay away from modeling things as one piece because it just seems messier. Is that actually wrong? Let's say you're making a FPS in 2020, should I actually be…