When you're learning to program, teachers will often tell you "The computer is dumb; it will do exactly what you say, and it can't read your intent." Implicitly, I think most people assume the same of compilers: they're dumb, and will translate an unconditional division as an unconditional division, even when it would crash.

Instead, compilers seemingly try to make insights into the intent or true meaning of your code, often inaccurately, and change it using ad-hoc unintuitive rules as justification (those being the rules of undefined behavior, and what assumptions the authors are and aren't allowed to make about your code). I know they aren't actually reading your intent; it's just the result of many, many passes that each make seemingly reasonable assumptions, which together may add up to a set of unreasonable ones.

> because that's what the programmer explicitly asked.

No, the programmer didn't explicitly ask for their check to be deleted, even with "unsafe" written. There is no request written anywhere. At most, you can say they implied they wanted it by invoking undefined behavior. Probably, though, they just made a mistake, or assumed that behavior was defined when it wasn't (e.g. they may have thought division corresponds to a hardware division instruction, modulo optimizations like right shifting for division by 2, constant folding division by 1, etc.)