Hacker News Reader

The thing is, CPUs are designed to execute machine code. Machine code isn't memory safe. Until we see memory safety implemented at the machine layer, memory safe languages will always be somehow treated as second-class citizens in the PL world.

View on HN

Well, we do have hardware memory safety, that's what the MMU is for. Clearly it's not being used enough. What we need to do is bring containerization down to the level of individual functions, so that the MMU can enforce encapsulation.

(I'm not sure whether this is a joke, a brilliant idea, or a piece of FP advocacy. You could probably get a CS graduate paper out of trying it, though)

Been done (damn, I'm channelling some of the arrogant beardy Usenix guys). Check the ScoutOS literature; one of the followons was a systems paper about a multiserver microkernel OS with memory boundaries between servers. Wherein we learn once again that it kills performance. (Dammit. Somebody stop my arrogant ass.)

MMU and MPUs had been around for ages. More exotic, tagged memory architectures also existed in the past.

Reply on news.ycombinator.com