Oracle SQL: how to enforce only one value may be 'checked'? - sql

Question

So I have a table in a database which contains the column "SELECTED". The values in this column can only be "CHECKED" or "UNCHECKED". I would like to enforce "CHECKED" can only be used once (like a radiobutton) through a PL/SQL trigger, though I cannot think of how to do this.
First, the idea (in case it didn't become clear):
Initial table "dummy":
ID | SELECTED
--------------
1 | 'UNCHECKED'
2 | 'CHECKED'
3 | 'UNCHECKED'
Then, I execute this query:
UPDATE dummy
SET SELECTED = 'CHECKED'
WHERE ID = 3;
Through a PL/SQL trigger, I'd like to have my table "dummy" to look like this after the execution:
ID | SELECTED
--------------
1 | 'UNCHECKED'
2 | 'UNCHECKED'
3 | 'CHECKED'
I hope you get the idea. I myself have tried to solve this, without success. I came up with the following code:
CREATE OR REPLACE TRIGGER DUMMY_ONE_CHECKED
AFTER INSERT OR UPDATE ON DUMMY
FOR EACH ROW
DECLARE
v_checked_is_present DUMMY.SELECTED%TYPE;
BEGIN
SELECT SELECTED
INTO v_checked_is_present
FROM DUMMY
WHERE SELECTED = 'CHECKED';
IF v_checked_is_present IS NOT NULL THEN
UPDATE DUMMY
SET SELECTED = 'UNCHECKED'
WHERE SELECTED = 'CHECKED';
UPDATE DUMMY
SET SELECTED = 'CHECKED'
WHERE ID = :NEW.ID;
END IF;
END;
However, I get the errors ORA-04091, ORA-06512 and ORA-04088 with the following message:
*Cause: A trigger (or a user defined plsql function that is referenced in
this statement) attempted to look at (or modify) a table that was
in the middle of being modified by the statement which fired it.
*Action: Rewrite the trigger (or function) so it does not read that table.
Clearly, this is not the right solution. I wonder how I could accomplish what I would like to do (if possible at all)?
Thank you in advance!

Answer 1

I would not design it that way. The database should enforce the rules, not automatically attempt to fix violations of them.
So, I'd enforce that only one row can be CHECKED at a time, like this:
CREATE UNIQUE INDEX dummy_enforce_only_one ON dummy ( NULLIF(selected,'UNCHECKED') );
Then, I'd make it the responsibility of calling code to deselect other rows before selecting a new one (rather than trying to have a trigger do it).
I know that doesn't answer the text of your question, but it does answer the title of your question: "how to enforce only one value..."

Answer 2

I'm not sure a trigger is the best approach to this problem. The trigger needs to update all the records for every update -- even worse, the rows are in the same table leading to the dreaded mutating table error.
How about a different table structure altogether? The idea is just to keep track of the last time something was "checked" and then use the maximum timestamp:
create table t_dummy (
id int,
checkedtime timestamp(6)
);
create view dummy as
select t_dummy.id,
(case when checkedtime = maxct then 'CHECKED' else 'UNCHECKED') as selected
from t_dummy cross join
(select max(checktime) as maxct from t_dummy) x;
This should be simpler to implement than a trigger.

Answer 3

One way to implement this is to use a COMPOUND TRIGGER. A compound trigger is one which has code which fires at each of the possible triggering points (BEFORE STATEMENT, BEFORE ROW, AFTER ROW, and AFTER STATEMENT). Let's look at how to handle your requirement:
CREATE OR REPLACE TRIGGER DUMMY_CHECKED_TRG
FOR INSERT OR UPDATE ON DUMMY
COMPOUND TRIGGER
TYPE NUMBER_TABLE IS TABLE OF NUMBER;
tblDUMMY_IDS NUMBER_TABLE;
BEFORE STATEMENT IS
BEGIN
tblDUMMY_IDS := NUMBER_TABLE();
END BEFORE STATEMENT;
AFTER STATEMENT IS
BEGIN
IF tblDUMMY_IDS.COUNT > 0 THEN
UPDATE DUMMY d
SET d.SELECTED = 'UNCHECKED'
WHERE d.ID <> tblDUMMY_IDS(tblDUMMY_IDS.LAST) AND
d.SELECTED = 'CHECKED';
END IF;
END AFTER STATEMENT;
AFTER EACH ROW IS
BEGIN
-- If the new value of `SELECTED` on this row is 'CHECKED'
-- save the ID of the row in tblDUMMY_IDS
IF :NEW.SELECTED = 'CHECKED' THEN
tblDUMMY_IDS.EXTEND;
tblDUMMY_IDS(tblDUMMY_IDS.LAST) := :NEW.ID;
END IF;
END AFTER EACH ROW;
END TABLE1_NUM_TRG;
In the BEFORE STATEMENT portion of the trigger we just allocate a table (variable length collection object) to hold ID values. This portion of the trigger is executed once, before any rows have been processed by the trigger.
In the AFTER EACH ROW section of the trigger we look at the SELECTED field of the row, and if it's 'CHECKED' we save its ID value in the table we allocated earlier.
The AFTER STATEMENT section of the trigger is where the real work gets done - and it's only a single SQL statement. The reason we defer the real work of the trigger until the AFTER STATEMENT section is because code which executes here will not raise the dreaded "MUTATING TABLE" exception. What we do is we take the last ID value which we found was associated with a row which had SELECTED = 'CHECKED'. This is the row which we want to remain CHECKED - every other row in the table should be UNCHECKED. So we execute an UPDATE statement, saying in effect "set SELECTED to 'UNCHECKED' on every row in the table whose ID is not the one we've got, and whose current value of SELECTED is CHECKED". Normally this will only update one row - but it will handle the case where in a single SQL statement sets a bunch of rows to CHECKED.
I believe compound triggers became available in 10g, so if you're on that version of Oracle or later you should be good.
Hope this helps.
Best of luck.

Answer 4

As I understand, you want to have only one row in the whole table, which could contain CHECKED value. But your way will not work.
I've just invented a new way how to do this. Maybe, it is a bit complicated way. Change your selected column type to number, and fill it with consequent numbers (for example, with sequence). Then consider column with maximal value as "selected". This gives you a lot of advantages: to change selected row you just need to take next value from a sequence and put it in desired row (you don't need to update all rows), you need only one query for that, and you never meet mutation problem. Disadvantages - quite hard to get selected row and impossible (hard) to "deselect all".

Answer 5

Alternative approach using table with one row only (enforced with PK). The `BUTTON_ID contains the ID of the selected button (1-3 or NULL if all buttons are un-checkedd). The button per row result is presented in a view.
create table button
(
id number primary key check (ID in (1)),
button_id number check (button_id in (1,2,3))
);
create view v_button as
with r3 as (select rownum button_id from dual connect by level <= 3)
select
case when button.button_id = r3.button_id then 'SELECTED' else 'UNSELECTED' end as button_code
from r3 cross join button
;
Initialize with
insert into button values(1,1);
gives
select * from v_button;
BUTTON_CODE
-----------
SELECTED
UNSELECTED
UNSELECTED
Switch simple with an update statement:
update button set button_id = 3;
gives
BUTTON_CODE
-----------
UNSELECTED
UNSELECTED
SELECTED
To de-select all simple set to NULL
update button set button_id = NULL;
BUTTON_CODE
-----------
UNSELECTED
UNSELECTED
UNSELECTED

Answer 6

Would be easier if you knew what needed to be unchecked. But if you can't, then unckeck everything.
UPDATE dummy
SET SELECTED = 'UNCHECKED';
Then check the one that you know should be checked.
UPDATE dummy
SET SELECTED = 'CHECKED'
WHERE ID = 3;
Why not use a boolean for this?
--EDIT-- (boolean example)
UPDATE dummy
SET SELECTED = 0;
Then check the one that you know should be checked.
UPDATE dummy
SET SELECTED = 1
WHERE ID = 3;