Re: master check fails on Windows Server 2008

Marina Polyakova <m(dot)polyakova(at)postgrespro(dot)ru> writes:
> Hello, hackers! I got a permanent failure of master (commit
> 2a41507dab0f293ff241fe8ae326065998668af8) check on Windows Server 2008.
> Regression output and diffs as well as config.pl are attached.

Weird. AFAICS the cost estimates for those two plans should be quite
different, so this isn't just a matter of the estimates maybe being
a bit platform-dependent. (And that test has been there nearly a
year without causing reported problems.)

To dig into it a bit more, I tweaked the test case to show the costs
for both plans, and got an output diff as attached. Could you try
the same experiment on your Windows box? In order to force the choice
in the other direction, you'd need to temporarily disable enable_sort,
not enable_hashagg as I did here, but the principle is the same.

regards, tom lane