I think the point is that in the cases where you can prove it's not possible, you can optimize it out. This is functionally the same, because if it's incapable of panicing because of out of bounds, you don't need to check for it. I think what you likely get in practice is something like:
func copy(src []int, dst []int) {
if (len(src) <= len(dst) {
for i := range(src) {
dst[i] = src[i]
} else {
for i := range(src) {
assert(i < len(dst)) // subscript check
dst[i] = src[i]
}
}
}
That way you still get panics in the case where they are applicable, but not in the cases where you can safely abstract the bounds check away (because a panic for that reason should not be possible). Although this is likely at the code generation level, so you can consider the if statement to be a sort of macro branching on what code is actually generated.