I remember having this realisation about Mir, but only after we collectively ran it off the cliff wall. The main reason everyone piled on Mir was that it was thought that Canonical would be priming Linux desktop for fragmentation with two competing standards.
But in fact, Mir was providing a solution to the fragmentation Wayland was bringing. Now we have 3, 4, 5 Mir-s, all with slight incompatibilities. Want a feature? Better hope all of them decide to implement the extension after someone proposes it. We know how well that worked in the past.
This is also ironic because the detractors of Xorg constantly talked about the issues with Xorg extensions and how many of them there were. But I never really had to look up which extensions Xorg supported, while I have had to do that with Wayland compositors.
Yeah OpenCASCADE is amazing because it's the only real geometry kernel that's open source. There's a few smaller ones like solvespace, but they're really more like toys. It's like the Linux of the CAD world.
Writing a geometry kernel is a monumental task, not unlike writing a real os kernel or a modern web engine. I've seen people just lay the basic foundations of a kernel as their PhD thesis. Most of the commercial ones were written decades ago and are still being worked on - the big ones are Parasolid ACIS, ShapeManager, CGM. The last one would maybe be considered a newcomer cause it's only 15-20 years old.