Thread

Re: BUG #19340: Wrong result from CORR() function

Tom Lane <tgl@sss.pgh.pa.us> — 2025-12-03T00:17:32Z

I tried the attached realization of your idea, and it does seem
very marginally faster than what I did; but it's like a 1.8%
slowdown instead of 2%.  Might be different on a different
machine of course.  But I think we should choose on the basis
of which semantics we like better, rather than such a tiny
performance difference.

I'm coming around to the conclusion that your way is better,
though.  It seems good that "any NaN in the input results in
NaN output", which your way does and mine doesn't.

			regards, tom lane