Luca Ferrari <fluca1...@gmail.com> writes:
> The ORDER BY rejects non existent columns (right) but accepts the
> table itself as an ordering expression.
As others have noted, this is basically taking the table name as a
whole-row variable, and then sorting per the rules for composite
types. I write to point out that you can often get some insight into
what the parser thought it was doing by examining the reverse-listing
for the query. The simplest way to do that is to create a view and
examine the view:
regression=# create view v as
regression-# select * from t order by t;
CREATE VIEW
regression=# \d+ v
View "public.v"
Column | Type | Collation | Nullable | Default | Storage | Description
--------+---------+-----------+----------+---------+---------+-------------
v | integer | | | | plain |
View definition:
SELECT t.v
FROM t
ORDER BY t.*;
The form "t.*" is a more explicit way to write a whole-row variable.
(IIRC, accepting it without "*" is a PostQUEL-ism that we've never
got rid of. I think that with "*", there's at least some support
for the concept in the SQL standard. But I'm insufficiently
caffeinated to want to go digging for that.)
regards, tom lane
