Ah, I think he’s claiming that the rendered geometry is not optimized for collision needs, and he’s right too.
Accordingly, the triangulation factor worsens the calculation of collisions even more, which is also true. The textbook’s conclusions that visible geometry should not be used for collision are convincing.
Which ultimately creates uncertainty and the need to choose a tool to create an optimized collision file in the editor. However, there is currently no solution for this.
I’m not convinced that this is so. The actual post reads to me otherwise (see for example that last sentence that I quoted above), and they previously posted that “[t]he custom collision shapes work exactly like regular geometry”.
I’m inclined to agree with you: I suspect that there’s more optimisation being done behind the scenes to make collisions with dedicated collision geometry more efficient than they would be with geometry intended for rendering.
(And indeed, I’ve warned others away from collision with visible geometry in the past, as I recall.)
Hence one of my uncertainties in this thread, I think: that if there had been a way to modify collision geometry, it might impact performance either by overriding that optimisation or by incurring that the optimisation happen every frame.
(Which is a moot point now, let me note, as we’ve established that there doesn’t seem to be a way to modify collision geometry on-the-fly.)