Even though this is not what this article is about, quite a few comments suggest better ways to compute the midpoint of two doubles, where "better" is supposed

Even though this is not what this article is about, quite a few comments suggest better ways to compute the midpoint of two doubles, where "better" is supposed to mean "such that unnecessary overflow is avoided". What I haven't seen is the suggestion 0.5*x + 0.5*y Yes, we now have two multiplications, but floating point multiplication is considerably faster than division.