I had the same basic thought while reading this: the only alternative here is to interpret this sort of undefined behavior in a "different" way (such as the way the K&R originally interpreted it) and leave it up the programmer to intuit what other assumptions the compiler is making on their behalf. Like... I can assume that the compiler will interpret:

    if (i++ < i)

as

    mov ax, i
    inc ax
    cmp ax, i
    jge else

... but that's still me making a (reasonable?) assumption. The C spec essentially just says "don't program that way".