The languages we're discussing here are high level - like C - and so they don't have addresses like the machine code, they have pointers. With pointers we're back to the compiler needing to know if they're valid - if you delay until runtime you lose not safety but performance before the compiler optimises based on knowing whether values can be modified via a pointer and if that's only decided at runtime we cannot perform these optimisations so now our program is slower or bigger or both.