|
Item_func_sec_to_time::get_date() executes val_str() to print warnings. So value methods for args[0] are executed two times:
1. val_int() or val_decimal() in get_deconds(), to get the "sec + sec_part" pair.
2. val_str(), to generate a warning, in case of overflow.
This looks wrong, especially for functions with side effect.
Example 1:
I run this script:
SET @a=10000000;
|
SELECT SEC_TO_TIME(@a:=@a+1);
|
SELECT @a;
|
It returns 838:59:59 from the first SELECT, with a warning as expected. But then the second SELECT returns a wrong result:
+----------+
|
| @a |
|
+----------+
|
| 10000002 |
|
+----------+
|
The expected result should be 10000001.
Example 2:
SELECT SEC_TO_TIME(LAST_VALUE(SLEEP(0.33),10000000));
|
+-----------------------------------------------+
|
| SEC_TO_TIME(LAST_VALUE(SLEEP(0.33),10000000)) |
|
+-----------------------------------------------+
|
| 838:59:59 |
|
+-----------------------------------------------+
|
1 row in set, 1 warning (0.66 sec)
|
Notice, it says 0.66 sec in the query statistics. The expected timing should be 0.33 sec, as in SLEEP.
Example 3:
DROP TABLE IF EXISTS t1;
|
DROP FUNCTION IF EXISTS f1;
|
CREATE TABLE t1 (a TEXT);
|
DELIMITER $$
|
CREATE FUNCTION f1() RETURNS INT
|
BEGIN
|
INSERT INTO t1 VALUES ('f1 was called');
|
RETURN 10000000;
|
END;
|
$$
|
DELIMITER ;
|
SELECT SEC_TO_TIME(f1());
|
SELECT * FROM t1;
|
+---------------+
|
| a |
|
+---------------+
|
| f1 was called |
|
| f1 was called |
|
+---------------+
|
Notice, the INSERT happened two times. The expected behavior would be to do INSERT only once.
|