Florian Wunderlich <fwunderlich@devbrain.de> writes:
> So can I be sure that every non-aggregate SELECT on tables joined with
> unique indexes works, independent of the WHERE or ORDER BY?
I would think that backwards scan on a join mostly doesn't work, but
haven't tried it in any detail.
A plan with a SORT node at the top *will* work, no matter what's under
the SORT, so ORDER BY might mask problems in some cases.
> Is anybody working on implementing this functionality?
Not me... although I have thought about at least adding enough code to
report an error in the cases that will give wrong answers.
regards, tom lane