INFORMIX SERVER / 5725A3900 / B70 - IDS 11.70

A query of following form might return more rows with an
additional EXISTS condition than without if the correlated
subquery returns more than one row for certain correlations:

SELECT 
   FROM A, B
 WHERE A.join_col = B.join_col
     AND 
     AND EXISTS (SELECT C.col_X from C WHERE C.join_col =
A.join_col AND );

This even would be true when adding UNIQUE to the subquery.
A '(Skip Duplicate)' seems to be of no effect.

The result set would be the same as for

SELECT 
   FROM A, B, C
 WHERE A.join_col = B.join_col
     AND A.join_col = C.join_col
     AND 
     AND ;

****************************************************************
* USERS AFFECTED:                                              *
* Users of 12.10.xC1 through 12.10.xC6.                        *
****************************************************************
* PROBLEM DESCRIPTION:                                         *
* A query with correlated subquery may produce incorrect       *
* resultS under the following conditions:                      *
*   - query contains an EXISTS or '= ANY' corrected subquery   *
*   - subquery has a single table, with an index on the join   *
* column of the subquery table (joining with a main query      *
* table)                                                       *
*   - a skip-duplicate index scan is chosen by the optimizer   *
* to scan the subquery table                                   *
*   - there are multiple (> 1) main query tables               *
*   - the optimizer chose a hash join between 2 main query     *
* tables (both after accessing the subquery table)             *
*   - the skip-duplicate index scan appears on the build side  *
* of the hash join, or if it appears on the probe side of the  *
* hash join, then hash join overflowed                         *
****************************************************************
* RECOMMENDATION:                                              *
* Update to IBM Informix Server 12.10.xC7                      *
****************************************************************

Use IN clause instead where possible.

not known / see Local fix

Date  - problem reported    :
Date  - problem closed      :
Date  - last modified       :

09.05.2016
29.06.2016
29.06.2016

12.10.xC7