MySQL group by all columns except one - mysql

Question

I'm looking for a (cleaner?) way to do the following:
Let's say I have a table, main, with ~15 columns that looks something like this, with one row per id:
main:
id start end col4 ... col15
666 2014-01-01 2014-06-30 ... ... ...
1234 2015-03-05 2015-05-02 ... ... ...
9876 2014-09-01 2015-01-01 ... ... ...
...(etc)
Then I have another table, events, which may have 0, 1, or many rows per id:
events:
id date code
666 2014-01-20 "code_a"
1234 2015-05-01 "code_b"
666 2014-01-25 "code_c"
666 2014-02-09 "code_z"
... (etc)
and finally I have a table, codes, which has one row per code, giving a description for the code as well as a type (0,1, or 2):
codes:
code desc type
"code_a" "something" 0
"code_b" "somethn else" 1
"code_c" "another thing" 0
"code_d" "one more" 2
(no code z)
and what I want as a result is main's 15 columns plus three additional columns which contain comma separated lists of event codes which happened between the start and end dates for that id by type (first column is type 0, second type 1, third type 2), so:
id start end ... col15 type_0 type_1 type_2
666 2014-01-01 2014-06-30 ... ... "code_a,code_c"
1234 2015-03-05 2015-05-02 ... ... "code_b"
...(etc)
my solution is
select m.*
, group_concat(c0.code) as type_0
, group_concat(c1.code) as type_1
, group_concat(c2.code) as type_2
from main m
left join events e on m.id = e.id and e.date between m.start and m.end
left join codes c0 on c0.code = e.code and c0.type = 0
left join codes c1 on c1.code = e.code and c1.type = 1
left join codes c2 on c2.code = e.code and c2.type = 2
group by m.id
, m.start
, m.end
, m.col4
, m.col5
, m.col6
, m.col7
, m.col8
, m.col9
, m.col10
, m.col11
, m.col12
, m.col13
, m.col14
, m.col15
But to me that's pretty nasty looking. Is there a more elegant way to do this (especially avoiding the 15 columns listed in the group by)?

Answer 1

In MySQL, you can just use GROUP BY m.id. Unless you enable the ONLY_FULL_GROUP_BY option, it allows you to use non-aggregate columns that aren't in the GROUP BY clause. This could produce unprectable results if you selected columns that were not uniquely identified by the grouping column, but that's not the case here -- you're grouping by a column that's the unique ID for the m table, and all the non-aggregate columns are from that same table.
In strict SQL, you would have to do it by doing the GROUP_CONCATs in a subquery, which you then join with the main table.
SELECT *
FROM (SELECT m.id,
, group_concat(c0.code) as type_0
, group_concat(c1.code) as type_1
, group_concat(c2.code) as type_2
FROM main m
left join events e on m.id = e.id and e.date between m.start and m.end
left join codes c0 on c0.code = e.code and c0.type = 0
left join codes c1 on c1.code = e.code and c1.type = 1
left join codes c2 on c2.code = e.code and c2.type = 2
GROUP BY m.id
) t1
JOIN main m ON t1.id = m.id

Answer 2

With the "one row per id" specification, you can take advantage of the MySQL extension to the GROUP BY which allows you to include non-aggregates in the SELECT list. The only change required to your query would be to just
GROUP BY m.id
Other databases would throw an error with that. We can get MySQL to throw an error too, if we include ONLY_FULL_GROUP_BY in the sql_mode for the session.
Another alternative would be to avoid a GROUP BY operation on m, using an inline view. You still need to do a GROUP BY, but you can do that in the inline view, where the other columns from main aren't returned, we only return the unique id value. We need that for the join in the outer query.
Also seems like you only need one join to the codes table; you could use a conditional test inside the GROUP_CONCAT to conditionally return the value of the code.
For example:
SELECT m.*
, g.type_0
, g.type_1
, g.type_2
FROM main m
LEFT
JOIN ( SELECT a.id
, GROUP_CONCAT(IF(c.type=0,c.code,NULL)) AS type_0
, GROUP_CONCAT(IF(c.type=1,c.code,NULL)) AS type_1
, GROUP_CONCAT(IF(c.type=2,c.code,NULL)) AS type_2
FROM main a
LEFT
JOIN events e
ON e.id = a.id
AND e.date BETWEEN a.start AND a.end
LEFT
JOIN codes c
ON c.code = e.code
AND c.type IN (0,1,2)
GROUP BY a.id
) g
ON g.id = m.id
I'm not sure that either of those qualifies as "a more elegant way" or not. (Both of these depend on the id column being UNIQUE in main. The second query also relies on id being non-NULL.)
You might want to consider adding an ORDER BY inside the GROUP_CONCAT, for a more deterministic result. It's also possible to include the DISTINCT keyword inside the GROUP_CONCAT, if there's no reason to return "duplicate" values of code in the list, e.g.
GROUP_CONCAT(DISTINCT IF(c.type=0,c.code,NULL) ORDER BY 1)\
Also be aware that the maximum length of the value returned from GROUP_CONCAT is limited to group_concat_max_len.

Answer 3

Another shorter version would be like below by getting the grouping first and then join with it.
select m.*
, XX.type_0
, XX.type_1
, XX.type_2
from main m
left join events e on m.id = e.id and e.date between m.start and m.end
left join (
select code, GROUP_CONCAT(case when type = 0 then code else null end SEPARATOR ', ') AS type_0,
GROUP_CONCAT(case when type = 1 then code else null end SEPARATOR ', ') AS type_1,
GROUP_CONCAT(case when type = 2 then code else null end SEPARATOR ', ') AS type_2
from codes
group by <some_column> )XX ON XX.code = e.code;