Ok, I found my mis-understanding and better understand where you are all coming from now; I apparently had the usage of NULL flip-flopped.
Taking pg_tablespace as an example. Its "spcacl" column produces NULL for default privileges and '{}'::text[] for empty privileges.
Thus, today:
empty: array_to_string('{}'::text[], '\n') produces an empty string
default: array_to_string(null, '\n') produces null which then was printed as a hard-coded empty string via forcibly changing \pset null
I was thinking the two cases were reversed.
My proposal for changing printACLColumn is thus:
case when spcacl is null then ''
when cardinality(spcacl) = 0 then '(none)'
else array_to_string(spcacl, E'\\n')
end as "Access privileges"
In short, I don't want default privileges to start to obey \pset null when it never has before and is documented as displaying the empty string. I do want the empty string produced by empty privileges to change to (none) so that it no longer is indistinguishable from our choice of presentation for the default privilege case.
Mechanically, we remove the existing \pset null for these metacommands and move it into the query via the added CASE expression in the printACLColumn function.
This gets rid of NULL as an output value for this column and makes the patch regarding \pset null discussion unnecessary. And it leaves the existing well-established empty string for default privileges alone (and changing this is what I believe Tom is against and I agree on that point).
David J.
