Okay, I've been able to reproduce the problem here.  Looks like the
eqjoinsel math is not wrong, exactly, but small roundoff errors are
causing the logic to do unreasonable things.  I think
get_att_numdistinct needs to round its result to an integer, and
probably there needs to be some clamping of computed probabilities to
the 0..1 range (otherfreq1 is coming out about -4.4703483581542969e-08
in this example, which should be clamped to 0).

Will have a fix late tonight or tomorrow.

Thanks for the example case!

			regards, tom lane

Responses

• Re: bug in join? at 2002-01-03 00:12:53 from Laurette Cisneros
• Re: bug in join? at 2002-01-03 04:05:23 from Tom Lane

pgsql-hackers by date

Next:	From: Tom Lane	Date: 2002-01-03 00:05:17
	Subject: Re: Bulkloading using COPY - ignore duplicates?
Previous:	From: Bruce Momjian	Date: 2002-01-02 23:40:20
	Subject: Re: Bulkloading using COPY - ignore duplicates?