At least CI here is happy, as much that means 🙂 Now I'm recompiling Julia on JuliaLang/julia@3c7bc45 with this change, to see if this is enough to fix compilation failure mentioned at #278 (comment).

It looks like feholdexcept() is also responsible for initializing fenv_t env, even if there are no actual hardware floating-point exception traps, just as on RISC-V. Instead of initializing env in the first place, can you try implementing fenv_t loading in feholdexcept(), which is a no-op on RISC-V?