ALTER TABLE DROP COLUMN may leave triggers invalid even if they are not using
the column getting dropped.
---------------------------------------------------------------------------------------------------------
Key: DERBY-4984
URL: https://issues.apache.org/jira/browse/DERBY-4984
Project: Derby
Issue Type: Bug
Components: SQL
Affects Versions: 10.7.1.1, 10.6.2.1, 10.6.1.0, 10.5.3.0, 10.5.2.0,
10.5.1.1, 10.4.2.0, 10.4.1.3, 10.3.3.0, 10.3.2.1, 10.3.1.4
Reporter: Mamta A. Satoor
While doing testing for DERBY-4887, I found a case where ALTER TABLE DROP
COLUMN will leave triggers in invalid state even if those triggers are not
using the column getting dropped. eg
CREATE TABLE tab (
element_id INTEGER NOT NULL,
altered_id VARCHAR(30) NOT NULL
);
insert into tab values(1,'aa');
-- Create a trigger against the table
CREATE TRIGGER mytrig
AFTER UPDATE OF altered_id ON tab
REFERENCING NEW AS newt OLD AS oldt
FOR EACH ROW MODE DB2SQL
SELECT newt.altered_id from tab;
--Drop the first column in the table. This will cause the column positions to
be recalculated within the table
alter table tab drop column element_id;
--mytrig is still looking for column altered_id at position 2 but drop column
has changed it's position within the table to 1
update tab set altered_id='bb';
As shown in the example above, table "TAB" only has 2 columns. The trigger
"MYTRIG" uses the 2nd column in it's trigger action through the REFERENCING
clause. During trigger action sql parsing, every column referenced through
REFERENCING clause gets transformed into a reference to the column through it's
column position in the trigger table(this change to look for columns based on
their column positions rather than the name went in as revision 397959 with
following commit comments DERBY-1258 Change the generated code for a new/old
column in a row trigger to access columns by position and not name to avoid the
case-insensitive name lookup specified by JDBC.) When in the script above, we
drop the column in position 1, the trigger "MYTRIG" ends up becoming invalid
because column being used in the trigger action is no more in column position 2.
One possible solution is to regenerate the SPSDescriptor associated with the
trigger action for all the triggers defined on the table whose column is
getting dropped. We could be little smarter and only regenerate the
SPSDescriptor for the triggers who use the REFERENCING clause. But we need to
do more testing to make sure that triggers without REFERENCING clause do not
get impacted by a drop of column which is not the last column of the table.
This optimization of recognizing the right triggers may not be worth it since
performance may not be that big a criteria for an ALTER TABLE DROP COLUMN which
should be a rare operation in a production system.
An interim solution to this problem is obviously to drop and recreate the
triggers
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
