Re: non-deterministic error related to MIN/MAX optimization

Jeff Davis <pgsql@j-davis.com> writes:
> On Mon, 2008-08-25 at 22:26 -0400, Tom Lane wrote:
>> ... It's mostly accidental that there are any
>> variants that do work, I think.  Why would you want a SRF in a sort key?

> The following line was added to the regression tests:
> aggregates.sql:226:
> select max(unique2), generate_series(1,3) as g from tenk1 order by g desc;

Hmm ... by me, even, according to the CVS logs :-)

> I have no argument with what you say above. But one of my colleagues at
> Truviso was doing some experiments, and it was causing a regression
> failure here. I should have been more clear.

Please provide some more detail about those experiments.  The test case
hasn't been seen to fail in the buildfarm, AFAIR.

            regards, tom lane