Using the ALTER COLUMN SET DATA TYPE extended support

ALTER COLUMN SET DATA TYPE was present in the earlier versions of DB2 as well, supporting SMALLINT to INTEGER, INTEGER to BIG, REAL to DOUBLE, and BLOB(n) to BLOB(n+m) conversions; data types could not be cast to smaller data types. In DB2 9.7, the ALTER TABLE statement is extended to support all compatible types, from casting to small data types.

In some cases, data may be truncated upon altering the column data type such as DECIMAL to INTEGER. To avoid the data loss issues, DB2 9.7 scans the column data before the change and writes the error messages, such as overflow errors and truncation errors, into the notification log.

The column data type is set to a new data type only if there is no error reported during the column data scan phase.

Getting ready

To perform the ALTER COLUMN SET DATA TYPE action, the user needs to have one of the following authorizations on the object:

• ALTER privilege

• CONTROL privilege

• ALTERIN privilege on the schema

• DBADM authority

How to do it...

We can do it using ALTER COLUMN SET DATA TYPE as follows:

ALTER TABLE SALES ALTER COLUMN SALES SET DATA TYPE SMALLINT
ALTER TABLE EMPLOYEE ALTER COLUMN COMM SET DATA TYPE INTEGER

The preceding SQL statements try to change the data type from one to another.

The reason for the failure in the case of second ALTER COLUMN statement is because an MQT "ADEFUSR" is referring to the base table EMPLOYEE.

db2 "? SQL0270N"
21

A column cannot be dropped or have its length, data type, security, or nullability altered on a table that is a base table for a materialized query table.

We can change the COLUMN type in the base table as follows:

CREATE TABLE BTABLE (C1 INT, C2 INT);
CREATE VIEW v1 AS SELECT C1, C2 FROM BTABLE;
CREATE VIEW v2 AS SELECT C1, C2 FROM V1;
ALTER TABLE BTABLE ALTER COLUMN C1 SET DATA TYPE SMALLINT;
REORG TABLE BTABLE;
SELECT SUBSTR(OBJECTNAME,1,20) NAME, SQLCODE, SQLSTATE, \
OBJECTTYPE FROM SYSCAT.INVALIDOBJECTS WHERE OBJECTNAME IN ('V1','V2')
SELECT * FROM v2;

How it works...

The ALTER COLUMN SET DATA TYPE statement downcasts the data type INT to SMALLINT, which invalidates the views V1 and V2. Since we have AUTO_REVAL set to DEFERRED, the dependent objects become invalid until used.

As soon as we access the dependent objects after altering the column data type, objects become valid and you won't see them in the SYSCAT.INVALIDOBJECTS system catalog table.

There's more...

• Casting of VARCHAR, VARGRAHIC, BLOB, CLOB, and DBCLOB data types to types smaller than the current one will not scan the table for a compatibility check, as this casting is not supported by DB2 9.7

• In the case of range partitioning tables, the string data type cannot be altered if it's a part of the range partitioning key

• The identity column cannot be altered in DB2 9.7

• In the case of the database partitioning feature, if the altering column is a part of the distribution key, then the new data type must meet the following listed conditions:

  • Same data type as the current one

  • Same column length

  • FOR BIT DATA cannot be modified in the case of CHAR and VARCHAR