From: pgsql-general-owner@postgresql.org [mailto:pgsql-general-owner@postgresql.org] On Behalf Of Ken Tanzer Sent: Thursday, October 11, 2012 4:49 PM To: pgsql-general@postgresql.org Subject: [GENERAL] non-integer constant in ORDER BY: why exactly, and documentation?
Hi. I recently ran a query that generate the same error as this:
SELECT * FROM generate_series(1,10) ORDER BY 'foo'; ERROR: non-integer constant in ORDER BY LINE 1: SELECT * FROM generate_series(1,10) ORDER BY 'foo';
The query was generated by an app (and the result somewhat inadvertent), so it was easy enough to change and I'm not asking here about a practical problem.
I am curious though about why this "limitation" exists. I get that integer constants are reserved for sorting by column numbers. But if Postgres already knows that it's a non-integer constant, why not let it go through with the (admittedly pointless) ordering?
Also, I couldn't see that this was explictly mentioned in the documentation. The relevant pieces seemed to be:
Each expression can be the name or ordinal number of an output column (SELECT list item), or it can be an arbitrary expression formed from input-column values.
I'm not sure if it would do violence to something I'm missing, but would the following combined statement work for the documentation?
"Each expression can be the name or ordinal number of an output column (SELECT list item), or it can be an arbitrary expression. The expression can include column values--whether they appear in the SELECT output list or not. An expression may not, however, consist solely of a non-integer constant. And an integer constant will be interpreted as the ordinal number of an output column "
I would categorize this under “help people avoid shooting themselves in the foot”. A possible situation is that the user meant to use double-quotes to specify an identifier but instead used single quotes. Since a literal constant would not impact the sort order the planner should either discard it silently or throw an exception. The exception is preferred since the presence of a constant literal likely means whatever generated the query is broken and should be fixed.
The documentation tweak probably is overkill given the rarity of the issue and the fact the system generates an appropriate error message when it does occur.
