information_schema.referential_constraints broken?

Answering my own question.
Replacing original definition of _pg_keysequal (in information_schema):

CREATE FUNCTION _pg_keysequal(smallint[], smallint[]) RETURNS boolean
LANGUAGE sql IMMUTABLE -- intentionally not STRICT, to allow inlining
AS 'select $1 <@ $2 and $2 <@ $1';

with this one:

CREATE FUNCTION _pg_keysequal(smallint[], smallint[]) RETURNS boolean
LANGUAGE sql IMMUTABLE -- intentionally not STRICT, to allow inlining
AS 'select $1 operator(pg_catalog.<@) $2 and $2 operator(pg_catalog.<@) $1';

makes queries like:

select * from information_schema.referential_constraints;

against information_schema.referential_constraints work without errors.

Sorry, for the noise.
Igor Neyman

From: Igor Neyman
Sent: Friday, July 27, 2012 9:27 AM
To: pgsql-general@postgresql.org
Subject: information_schema.referential_constraints broken?

This query:

select * from information_schema.referential_constraints;

under PG 8.4.5 (Windows platform) produces this error message:

ERROR: operator is not unique: smallint[] <@ smallint[]
LINE 1: select $1 <@ $2 and $2 <@ $1
^
HINT: Could not choose a best candidate operator. You might need to add explicit type casts.
QUERY: select $1 <@ $2 and $2 <@ $1
CONTEXT: SQL function "_pg_keysequal" during inlining

********** Error **********

ERROR: operator is not unique: smallint[] <@ smallint[]
SQL state: 42725
Hint: Could not choose a best candidate operator. You might need to add explicit type casts.
Context: SQL function "_pg_keysequal" during inlining

I don't have more recent 8.4 releases to try it on.
It works fine on PG 9.1.3.

Did anyone else experienced this problem?

Regards,
Igor Neyman

Igor Neyman <ineyman@perceptron.com> writes:

Answering my own question.
Replacing original definition of _pg_keysequal (in information_schema):

CREATE FUNCTION _pg_keysequal(smallint[], smallint[]) RETURNS boolean
LANGUAGE sql IMMUTABLE -- intentionally not STRICT, to allow inlining
AS 'select $1 <@ $2 and $2 <@ $1';

with this one:

CREATE FUNCTION _pg_keysequal(smallint[], smallint[]) RETURNS boolean
LANGUAGE sql IMMUTABLE -- intentionally not STRICT, to allow inlining
AS 'select $1 operator(pg_catalog.<@) $2 and $2 operator(pg_catalog.<@) $1';

makes queries like:
select * from information_schema.referential_constraints;
against information_schema.referential_constraints work without errors.

This is known to happen pre-9.0 if you have contrib/intarray installed.

regards, tom lane