Halting problem

Yeah. The problem here is not the halting problem: the problem here is always finding an approximation that lets useful stuff past but isn't too complex to be worth it. The high time- or sometimes space-complexity of many algorithms needed by compilers is a routine cause of foul curses in developers of normal ahead-of-time compilers that don't run in kernel space: the approximations needed by something that runs in the kernel seem likely to be even harder.

(Really, it's a shame the verifier and JITters have to run in kernel space: this sort of work seems like the sort of thing far better done by an upcall to something in userspace, unprivileged but only installable by root -- only that would stop BPF from being used to protect the system *from* root, and it wouldn't help with the time problem at all since the kernel would still have to block the process installing the BPF to wait for verification/JITing anyway...)

It just makes me icky seeing that people are basically writing compilers that run in kernelspace. Surely there must be a better way.