On Sat, 6 Jan 2024 at 16:22, David G. Johnston <david.g.johns...@gmail.com> wrote: > On Sat, Jan 6, 2024 at 8:38 AM Geoff Winkless <pgsqlad...@geoff.dj> wrote: >> because when gp_conc is 0, it should be ordering by the concat() value. > > Something does seem off here with the interaction between grouping sets and > order by. > I'm inclined to believe that using grouping in the order by simply is an > unsupported > concept we fail to prohibit.
That's disappointing. > You can get the desired result with a much less convoluted order by clause - > so long as you understand where your nulls are coming from - with: > ORDER BY > n nulls first , x nulls first Ahh, well, yes, that's fine in this instance which (as you may remember) was a minimal example of the behaviour, but wouldn't be useful in the real-world situation, where we can have many potentially-conflicting grouping sets, each set needing to be ordered consistently internally. Geoff
