infinite loop optimized away unless function is exported

[numsim1415]

export fn square(num: u8) u8 {
             
             //while(true){} // would generate loop
             return eternity(123); 
}

pub fn eternity(bla:u8) u8 {
  while(true)// is optimized away completely
  {}
}

See asm output on goldbold.
As you said during live coding, it seems like the eternal loop is considered to have no side effects and is thus removed.

Expected behaviour would be to refuse compilation

• as there is no return of an u8
• as in case of eternity being declared as returning void, an implicit return; at the end of the function cannot be reached

[IridescentRose]

Isn't this a long time LLVM bug present in other languages like C and C++ where an empty infinite loop results in no consequences and thus creates invalid programs? (for example there are many articles like https://stackoverflow.com/questions/59925618/how-do-i-make-an-infinite-empty-loop-that-wont-be-optimized-away). I guess the solution Zig would need for this specific case is adding in a nop op-code inside of the loop so LLVM says it "has an effect" and doesn't generate an invalid program/function.

[IridescentRose]

 export fn square(num: u8) u8 {          
    //while(true){}
    return eternity(123); 
}

pub fn eternity(bla:u8) u8 {
    while(true){
        asm volatile("nop");
    }
}

This code generates as intended as we can see on GodBolt.

 export fn square(num: u8) u8 {          
    //while(true){}
    return eternity(123); 
}

pub fn eternity(bla:u8) u8 {
    while(true){
        asm volatile("nop");
    }
}

This code generates as intended as we can see on GodBolt.

You probably have forgot to apply the flag -Drelease-safe, this happens to me as well on godbolt using Zig trunk