Re: patternsel() and histogram_selectivity() and the hard cutoff of 100

Gregory Stark <stark(at)enterprisedb(dot)com> writes:
> "Gregory Stark" <stark(at)enterprisedb(dot)com> writes:
>> So I had a thought about how to soften the controversial hard cutoff of 100
>> for the use of the histogram selectivity. Instead of switching 100% one way or
>> the other between the two heuristics why not calculate both and combine them.
>> The larger the sample size from the histogram the more we can weight the
>> histogram calculation. The smaller the histogram size the more we weight the
>> heuristic.

> Incidentally I hacked up a patch to do this:

Applied with revisions --- I thought it was better to let the caller of
histogram_selectivity make the decision about how to combine the
estimates, instead of hard-wiring the choice into that subroutine.

regards, tom lane