This reminds me of LLD’s approach of sharing an architecture but having separate ELF, COFF and Mach-O implementations. This made it much faster than gold which wasn’t actually slow.

Apparently sharing code isn’t necessarily a great idea.