Neil Conway <neilc@samurai.com> writes:
> On Mon, 2007-28-05 at 15:05 -0400, Tom Lane wrote:
>> Hmmm ... if you check the cvs history for those tests you might find
>> some evidence.
> "Modify regression tests to allow GEQ optimizer (order results).",
> according to the 1997 CVS commit from Thomas Lockhart that added the
> lines in question. Presumably that is no longer relevant.
Some digging in the CVS history shows that the parameter now known as
geqo_threshold didn't exist until here:
1997-05-31 22:55 momjian
* src/: backend/optimizer/path/allpaths.c,
include/optimizer/geqo.h: Enable GEQO for more than six tables,
from Martin.
so at the time Tom was griping about this, GEQO would indeed have
randomized the results for even two-table joins.
> Nothing
> unexpected happens if the disabling of GEQO is removed, so I'll apply
> the attached patch shortly barring any objections.
No objection here.
regards, tom lane