Re: [COMMITTERS] pgsql: Add STRICT to PL/pgSQL SELECT INTO, so exceptions

Tom Lane wrote:
> momjian@postgresql.org (Bruce Momjian) writes:
> > Add STRICT to PL/pgSQL SELECT INTO, so exceptions are thrown if more or
> > less than one row is returned by the SELECT, for Oracle PL/SQL
> > compatibility.
> 
> I've got a couple of problems with the error codes used by this patch.
> In the first place, you can't arbitrarily assign names to error
> conditions that are different from the standard spelling (see
> errcodes.sgml for why not: the standard spellings are what are
> documented).  In the second place, the spec clearly says that class 02

I saw this at the top of plerrcodes.h: * Eventually this header file should be auto-generated from errcodes.h * with
somesort of sed hackery, but no time for that now.  It's likely * that an exact mapping will not be what's wanted
anyhow...
 

so I figured we were supposed to map them.

> is warning conditions, not errors, so using ERRCODE_NO_DATA for an error
> is inappropriate.

Oh, I see that now:/* Class 02 - No Data --- this is also a warning class per SQL99 *//* (do not use this class for
failureconditions!) */#define ERRCODE_NO_DATA                     MAKE_SQLSTATE('0','2', '0','0','0')
 

> Where did you get those names from ... were they picked out of the air,
> or were they intended to be Oracle-compatible, or what?  Surely we

I pulled this from the Oracle documentation that I quoted earlier:

> >     When you use a SELECT INTO statement without the BULK COLLECT clause, it
> >     should return only one row. If it returns more than one row, PL/SQL
> >     raises the predefined exception TOO_MANY_ROWS.
> >
> >     However, if no rows are returned, PL/SQL raises NO_DATA_FOUND unless the
> >     SELECT statement called a SQL aggregate function such as AVG or SUM.
> >     (SQL aggregate functions always return a value or a null. So, a SELECT
> >     INTO statement that calls an aggregate function never raises
> >     NO_DATA_FOUND.)

Are those both errors in Oracle?  I assumed so.

> aren't trying to be Oracle-compatible at that level of detail (else
> we've doubtless got a huge number of other cases where we throw a
> different error than they do).

I thought it was nice to get as close as possible, but using a warning
code is clearly bad.

> Do we actually need different error codes for too few and too many rows?
> It looks to me like the only relevant standard error condition is
> CARDINALITY_VIOLATION, so either we throw CARDINALITY_VIOLATION for both
> cases or we invent nonstandard codes.

We could, and then suggest using ROW_COUNT to determine if there were
too few rows, or too many.

--  Bruce Momjian   http://candle.pha.pa.us EnterpriseDB    http://www.enterprisedb.com
 + If your life is a hard drive, Christ can be your backup. +