Re: 8.2 bug with outer join reordering

Jeff Davis <pgsql@j-davis.com> writes:
> On 8.1 this returns 1 record. On 8.2 this returns 100000. It appears to
> be applying the filter too soon, and then it does an outer join which
> violates the WHERE.

AFAICS the outer join reordering is perfectly legal --- the problem is
that the WHERE condition is being allowed to bubble down too far.
I can't reproduce it with less than four tables, so it's a pretty
weird corner case.  Apparently there's something wrong with
distribute_qual_to_rels' logic for determining qual placement, but
I'm not sure what yet ...

            regards, tom lane