Re: Aggregate ORDER BY patch

>>>>> "Tom" == Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:

>> Query-level DISTINCT shouldn't allow columns in the order by that
>> aren't in the select list because those columns _do not exist_ at
>> the point that ordering logically takes place (even though in the
>> implementation, they might).

>> This isn't the case for aggregate order by.

Tom> I entirely disagree. Why should the semantics of this
Tom> combination of ORDER BY and DISTINCT be different from what they
Tom> are at the query top level? We made other decisions about this
Tom> feature on the basis of making the two cases work alike, and I
Tom> don't think you've made an adequate argument for making them act
Tom> differently.

A case could possibly be made that the behaviour of DISTINCT at top
level is wrong, or at least less useful than need be.

Notice that there are cases where agg(distinct x order by x) is
nondeterministic while agg(distinct x order by x,y) is deterministic.
In my view that alone is a good argument for allowing it.

--
Andrew (irc:RhodiumToad)