Re: [HACKERS] Drop one-argument string_agg? (was Re: string_agg delimiter having no effect with order by)

Josh Berkus <josh(at)agliodbs(dot)com> writes:
>> Well, maybe we need to expend some more sweat on the error message then.
>> But this patch was still a prerequisite thing, because without it there
>> is no error that we can complain about.

> Yes, I'd say an addition to the HINT is in order *assuming* at that
> stage we can tell if the user passed an ORDER BY or not.

I was just looking at this, and realized I was mistaken earlier: the
error is issued in ParseFuncOrColumn, which already is passed the
agg_order list, so actually it's completely trivial to tell whether
a variant error message is appropriate. I suggest that we key it off
there being not just an ORDER BY, but an ORDER BY with more than one
element; if there's only one then this cannot be the source of
confusion.

Next question: exactly how should the variant HINT be phrased?
I'm inclined to drop the bit about explicit casts and make it read
something like

HINT: No aggregate function matches the given name and argument
types. Perhaps you misplaced ORDER BY; ORDER BY must appear after all
regular arguments of the aggregate.

regards, tom lane