Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> writes:
> On Sat, 16 May 2026 at 17:45, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> One simple change that might make things better is to compute
>> PG_RETURN_FLOAT8((Sy - Sx * (Sxy / Sxx)) / N);
>> on the theory that the sums of products are likely to both be large.
> Hmm, that isn't necessarily better.
True. We could do something like
(Sy - exp(log(Sx) + log(Sxy) - log(Sxx))) / N
but this'd require additional special-case code to deal with zero
or negative Sx and Sxy, so it's feeling rather tedious.
regards, tom lane