On Sat, Apr 23, 2016 at 10:39 AM, Geoff Winkless <pgsqladmin@geoff.dj> wrote:
> SELECT with no values is selecting a NULL, and two NULLs do not
> equate, so you would expect two rows.
This is precisely what I mean by wonky semantics: it makes no sense
for a nullary relation to be interpreted as selecting a NULL. A row
with no columns should have different semantics from a row with one
NULL column. Indeed it does in other contexts; note its correct
behavior as the neutral for the cross join operator:
postgres=# select * from (values (42)) as t(n) cross join (select) as dee;
n
----
42
(1 row)
postgres=# select * from (values (42)) as t(n) cross join (select
NULL) as not_dee;
n | ?column?
----+----------
42 |
(1 row)
The empty tuple should equal itself and its semantics should have
nothing at all to do with NULL, and thence I expect one row. I've
even been tempted to file a bug report, but I fear it may be brushed
off as pedantry.