There has been research on escape analysis, static memory management and all. Even not considering equivalence to halting problem, it has proven basically impossible to manage memory at compile time except very simple programs. This is a reason why rust ownership semantics are so complicated. And still anything complicated will use reference counting for many shared references. (Eg: Linux kernel).
But this guy is not even good at marketing he is doing -- some promises look very shady and comparisons are not apples to apples.
I see how big efforts zig and odin authors made, and the languages are impressive. V is a hyped piece from beginning and this makes the author seem more shadier.
I read about this or a similar approach before. I've got no idea how they think they'll pull it off without requiring rust like annotations or reducing expressive power to something below TM equivalence. Or "leak free" doesn't mean what I think it does.
Notice that after it's stable it will be on by default with a `-noautofree` option.