Reputation: 4989
Say I have a query which returns the following result:
| val | type | i |
----------------------
| 59 | 1 | 1 |
| 40 | 2 | 2 |
| 12 | 1 | 3 |
| 3 | 2 | 4 |
| 24 | 1 | 5 |
| 30 | 1 | 6 |
| 98 | 2 | 7 |
| 45 | 2 | 8 |
| 46 | 1 | 9 |
val = an arbitrary number
type = 1 or 2
i = auto increment column, just for reference
I want to perform the following subtraction:
A - B
A: Value of row of type '2'
B: Value of row above A of type '1' if it exists, but without crossing a row of type '2'
I really hope that made sense..
Some examples:
Row i=2 is of type 2. So the first row above it of type 1 is i=1.
So the subtraction will be: 40 - 59 = -19
Row i=7 is of type 2. So the first row above it of type 1 is i=6
So the subtraction will be: 98 - 30 = 68
Row i=8 is of type 2. But if we go above we cross a value of type 2.
So the subtraction should return 0, or be ignored, whichever is simplest.
In the end the result should return a column of values where type=1 and the subtracted value .
Eg:
| val | diff |
--------------
| 59 | -19 |
| 12 | -9 |
| 24 | 0 | ***
| 30 | 68 |
| 46 | 0 | ***
*** Return diff=0 if the subtraction was never made for this row.
I have the query for getting my initial table. I know how to do the subtraction (in the real case val is a date, but I'm keeping it simple for this example).
The only thing I have no idea how to do is perform logic based on different rows in MySQL. For example how can I find the nearest row based on some condition and then reference that row to perform some operation on it? Is there perhaps a way to reference the current 'i' and do an subrtaction on (row at i) minus (row at i-1) or something similar.
I've spent many hours on this already and I'm just about to give up and do it in php instead and take the performance hit. As a last resort I'm asking here if there's a way to do all this in mysql directly?
I don't need a full solution as this is pretty complex requirement but I will take any sort of advice or pointers you can give me. I'm not the best at MySQL so maybe I'm missing something simple.
Thanks.
Upvotes: 1
Views: 2639
Reputation: 10635
SELECT t1.val AS val, IFNULL( t2.val, t1.val ) - t1.val AS diff FROM my_table AS t1
LEFT JOIN my_table AS t2 ON( t2.type = 2 AND t2.i = t1.i + 1 )
WHERE t1.type = 1 ORDER BY t1.i;
Tested on your data, result:
| val | diff |
--------------
| 59 | -19 |
| 12 | -9 |
| 24 | 0 |
| 30 | 68 |
| 46 | 0 |
Upvotes: 2
Reputation: 338
Maybe you can do something like this:
select t1.val, (t1.val - t2.val) as diff
from table as t1, table as t2
where t1.i = t2.i + 1
and t1.type = 1
and t2.type = 2
to select with zeros you can do a hack like this:
select t1.val, (t1.val - t2.val)*(t2.type - t1.type) as diff
from table as t1, table as t2
where t1.i = t2.i + 1
and t1.type = 1
to get the last zero when there is one, join with:
select val, 0 as diff from
select val
from table
order by i desc
limit 1
Upvotes: 0
Reputation: 16037
What you need is a stored procedure with a cursor on your table. This should be a good starting pont.
CREATE PROCEDURE curdemo()
BEGIN
DECLARE done INT DEFAULT 0;
DECLARE val, type, prev_val, prev_type INT;
DECLARE cur1 CURSOR FOR SELECT val, type, i from your_table;
DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = 1;
OPEN cur1;
read_loop: LOOP
FETCH cur1 INTO val, tpye, i;
IF done THEN
LEAVE read_loop;
END IF;
-- your logic goes here, sorry, not sure I understood correctly, but roughly here it goes
IF type = 2 THEN
IF prev_type is not null and prev_type = 1 THEN
INSERT INTO result_table VALUES (prev_val, val - prev_val);
END IF;
END IF;
SET prev_val = val;
SET prev_type = type;
END LOOP;
CLOSE cur1;
END;
Upvotes: 0