"David G. Johnston" <david.g.johnston@gmail.com> writes: > On Thu, Jun 22, 2023 at 5:08 PM Tom Lane <tgl@sss.pgh.pa.us> wrote: >> * Personally I could do without the "empty" business, but that seems >> unnecessary in the tabular format; an empty column will serve fine.
> I disagree, but not strongly.
> I kinda expected you to be on the side of "why are we discussing a > situation that should just be prohibited" though.
I haven't formed an opinion yet on whether it should be prohibited. But even if we do that going forward, won't psql need to deal with such cases when examining old servers?
I haven't given enough thought to that. My first reaction is that using blank for old servers would be desirable and then, if allowed in v16+ server, "empty" for those.
That said, the entire grantor premise that motivated this doesn't exist on those servers so maybe \drg just shouldn't work against pre-v16 servers - and we keep the existing \du query as-is for those as well while removing the "member of" column when \du is executed against a v16+ server.