MySQL: splits row into multiple rows with join table? - mysql

Question

I have an existing table with the following structure
+-------------+-------+-------+-------+
| employee_id | val_1 | val_2 | val_3 | ...
+-------------+-------+-------+-------+
| 123 | A | B | C |
I want to change this single table into 2 tables - one which contains the values in seperate rows, and another with becomes a join table for this. For example, the above would be turned into this:
+-------------+--------+ +----+-------+
| employee_id | val_id | | id | value |
+-------------+--------+ +----+-------+
| 123 | 1 | | 1 | A |
+-------------+--------+ +----+-------+
| 123 | 2 | | 2 | B |
+-------------+--------+ +----+-------+
| 123 | 3 | | 3 | C |
+-------------+--------+ +----+-------+
What's the best SQL to use to convert the existing table into these 2 new tables? I can create the values table easy enough, but I'm not sure how to create the join table at the same times.

Answer 1

Something like this (psuedo-code only, sorry):
For each row in (SELECT employee_id, val_1, val_2, val_3 FROM existing_table)
{
for each val in (row.Values)
{
INSERT INTO new_values (val)
val_id = SELECT LAST_INSERT_ID();
INSERT INTO new_employees (employee_id, val_id);
}
}
There's probably a set-based way of doing this to avoid the loops... but sorry, I don't know what it is as like you, I'm not sure how to get the identity of the values table back into the parent employee table.
And also, while cursors are generally frowned on, this sort of one-off operation is exactly what they're designed for (ie I wouldn't recommend cursors for regular transaction or report processing, but for a re-structure of data.... why not?).

Answer 2

for the first result
`INSERT INTO new_val
SELECT emp_id, REPLACE(UPPER(column_name), 'VAL_', '') FROM
information_schema.COLUMNS ,
employee
WHERE TABLE_NAME = 'employee' AND TABLE_SCHEMA = 'myschema' AND column_name LIKE 'VAL_%'`;
using first result, populate the query and use it to insert into new table;
May be minor fine tuning required. Not tested
SELECT CONCAT('select ', new_val.number, ', VAL_', new_val.number, '
FROM employee, new_val
WHERE new_val.emp_id = employee.emp_id and new_val.number = ', val.number,
' union all' ) FROM
val ;

Answer 3

So here's what I ended up writing to do this. As the procedure name suggests, I was expecting there to be a more straightforward way of doing this!
CREATE PROCEDURE iWasHopingItWouldBeSimpler()
BEGIN
DECLARE loop_done BOOLEAN DEFAULT 0;
DECLARE emp_id BIGINT(20);
DECLARE val1 DECIMAL(19,2);
DECLARE val2 DECIMAL(19,2);
DECLARE val3 DECIMAL(19,2);
DECLARE emp CURSOR
FOR
SELECT employee_id, val1, val2, val3 FROM existing;
-- Declare continue handler
DECLARE CONTINUE HANDLER FOR SQLSTATE '02000' SET loop_done=1;
OPEN emp;
-- Loop through all rows
REPEAT
FETCH emp INTO emp_id, val1, val2, val3;
INSERT INTO new_values (value) VALUES(val1);
INSERT INTO new_join (employee_id, values_id) VALUES(emp_id, LAST_INSERT_ID());
INSERT INTO new_values (value) VALUES(val2);
INSERT INTO new_join (employee_id, values_id) VALUES(emp_id, LAST_INSERT_ID());
INSERT INTO new_values (value) VALUES(val3);
INSERT INTO new_join (employee_id, values_id) VALUES(emp_id, LAST_INSERT_ID());
-- End of loop
UNTIL loop_done END REPEAT;
CLOSE emp;
SET loop_done=0;
END;