Robert Haas <robertmhaas@gmail.com> writes:
> On Wed, Aug 31, 2011 at 6:22 AM, Peter Eisentraut <peter_e@gmx.net> wrote:
>> I liked the old one better. ;-)
> AFAICS, those plans are identical, except for a minor difference in
> the cost of scanning test2.
The point is that the estimate of the result size is worse in 8.4.8.
I am not, however, convinced that 8.4.7 was actually smarter ... it
may have been getting the right answer for the wrong reason.
regards, tom lane