Pavel Stehule <pavel.stehule@gmail.com> writes:
>>>> The same problem can be with custom aggregates :( so this syntax isn't
>>>> too robust.
BTW, I'm really not worried about that case. By the time someone is
advanced enough to have written their own multi-argument aggregate
definitions, they'll have absorbed the idea that the ORDER BY goes at
the end. What we need to accomplish here is just to not set traps at
the feet of novices using the feature for the first time. Which is
why I think it's sufficient to have a policy of not having built-in
aggregates that conflict in this way; I'm not proposing that we restrict
or discourage custom aggregates with optional arguments.
regards, tom lane