RISC-V never had a fragmentation problem, thanks to the profiles.

I wouldn't say it never had a problem, but the profiles are definitely a reasonable solution.

However even with profiles there are optional extensions and a lot of undefined behaviour (sometimes deliberately, sometimes because the spec is just not especially well written).

	▲	snvzz a day ago \| parent [-]
		The FUD keeps being brought up, but the solution here was in place before the potential issue could manifest. It started with G, later retroactively named RVA20 (with a minor extra extension that nobody ever skipped implementing), then RVA22 and now RVA23. All application processor implementations out there conform to a profile, and so do the relevant Linux distributions. Of course, in embedded systems where the vendor controls the full stack, the freedom of micromanaging which extensions to implement as well as the freedom to add custom extensions is actual value. The original architects of the ISA knew what they were doing.