Обсуждение: Strange behavior on non-existent field in subselect?

We're a little puzzled by this (apparently) strange behavior, and would
be curious to know what you folks make of it.  Thanks.

Ken

CREATE TABLE foo (
     foo_field   integer );

CREATE TABLE par(
     par_field   integer );

SELECT VERSION();

SELECT foo_field FROM par;
SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);
INSERT INTO foo VALUES (1);
SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);
INSERT INTO par VALUES (1);
SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);
/* One row for every foo record, provided at least one record in par */

Which (for us) yields the following output:

Chasers=> \i strangefield.sql
CREATE TABLE
CREATE TABLE
                                                 version
-------------------------------------------------------------------------------------------------------
  PostgreSQL 8.1.4 on i686-redhat-linux-gnu, compiled by GCC gcc (GCC)
3.4.4 20050721 (Red Hat 3.4.4-2)
(1 row)

psql:strangefield.sql:11: ERROR:  column "foo_field" does not exist
  foo_field
-----------
(0 rows)

INSERT 0 1
  foo_field
-----------
(0 rows)

INSERT 0 1
  foo_field
-----------
          1
(1 row)

On þri, 2006-10-17 at 15:58 -0700, Ken Tanzer wrote:
> We're a little puzzled by this (apparently) strange behavior, and would
> be curious to know what you folks make of it.  Thanks.

not sure exactly what you are referring to, but:
(rearranged quotes to group output with SQL)

> SELECT foo_field FROM par;
> psql:strangefield.sql:11: ERROR:  column "foo_field" does not exist

hopefully, no mystery here.


> SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);

if par is empty, then this SELECT will return 0 rows,
otherwise it is equivalent to SELECT foo_field from foo

>  foo_field
> -----------
> (0 rows)

foo is empty, so no rows returned

> INSERT INTO foo VALUES (1);
> SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);
>   foo_field
> -----------
> (0 rows)

par is empty, so the IN operator fails for the foo row

> INSERT INTO par VALUES (1);
> SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);
>   foo_field
> -----------
>           1
> (1 row)

when par contains at least one row, the subselect will
return foo_field once per row of par.
the IN operator will ignore duplicates, so the result
is the same for any number of rows in par greater than 0

gnari

Thanks for the response Ragnar.  I would have expected this query to
fail, since the sub-query doesn't work by itself:

>> SELECT foo_field FROM foo WHERE foo_field IN (SELECT foo_field FROM par);

But it obviously doesn't.  So does that subselect implicitly read as:

IN (SELECT foo_field FROM par,foo);

Thanks for your help!

Ken