If the user_resources (t1) was a 'normalized table' with one row for each user => resource combination then the query to get the answer would be as simple as just joining the tables together.
Alas, it is denormalized by having the resources column as a: 'list of resource id' separated by a ';' character.
If we could convert the 'resources' column into rows then a lot of the difficulties go away as the table joins become simple.
The query to generate the output asked for:
SELECT user_resource.user,
resource.data
FROM user_resource
JOIN integerseries AS isequence
ON isequence.id <= COUNT_IN_SET(user_resource.resources, ';') /* normalize */
JOIN resource
ON resource.id = VALUE_IN_SET(user_resource.resources, ';', isequence.id)
ORDER BY
user_resource.user, resource.data
The Output:
user data
---------- --------
sampleuser abcde
sampleuser azerty
sampleuser qwerty
stacky qwerty
testuser abcde
testuser azerty
How:
The 'trick' is to have a table that contains the numbers from 1 to some limit. I call it integerseries. It can be used to convert 'horizontal' things such as: ';' delimited strings into rows.
The way this works is that when you 'join' with integerseries, you are doing a cross join, which is what happens 'naturally' with 'inner joins'.
Each row gets duplicated with a different 'sequence number' from the integerseries table which we use as an 'index' of the 'resource' in the list that we want to use for that row.
The idea is to:
- count the number of items in the list.
- extract each item based on its the position in the list.
- Use
integerseries to convert one row into a set of rows extracting the individual 'resource id' from user.resources as we go along.
I decided to use two functions:
function that given a 'delimited string list' and an 'index' will return the value at the position in the list. I call it: VALUE_IN_SET. i.e. given 'A;B;C' and an 'index' of 2 then it returns 'B'.
function that given a 'delimited string list' will return the count of the number of items in the list. I call it: COUNT_IN_SET. i.e. given 'A;B;C' will return 3
It turns aout that those two functions and integerseries should provide a general solution to delimited items list in a column.
Does it work?
The query to create a 'normalized' table from a ';' delimited string in column. It shows all the columns, including the generated values due to the 'cross_join' (isequence.id as resources_index):
SELECT user_resource.user,
user_resource.resources,
COUNT_IN_SET(user_resource.resources, ';') AS resources_count,
isequence.id AS resources_index,
VALUE_IN_SET(user_resource.resources, ';', isequence.id) AS resources_value
FROM
user_resource
JOIN integerseries AS isequence
ON isequence.id <= COUNT_IN_SET(user_resource.resources, ';')
ORDER BY
user_resource.user, isequence.id
The 'normalized' table output:
user resources resources_count resources_index resources_value
---------- --------- --------------- --------------- -----------------
sampleuser 1;2;3 3 1 1
sampleuser 1;2;3 3 2 2
sampleuser 1;2;3 3 3 3
stacky 2 1 1 2
testuser 1;3 2 1 1
testuser 1;3 2 2 3
Using the above 'normalized' user_resources table, it is a simple join to provide the output required:
The functions needed (these are general functions that can be used anywhere)
note: The names of these functions are related to the mysql FIND_IN_SET function. i.e. they do similar things as regards string lists?
The COUNT_IN_SET function: returns the count of character delimited items in the column.
DELIMITER $$
DROP FUNCTION IF EXISTS `COUNT_IN_SET`$$
CREATE FUNCTION `COUNT_IN_SET`(haystack VARCHAR(1024),
delim CHAR(1)
) RETURNS INTEGER
BEGIN
RETURN CHAR_LENGTH(haystack) - CHAR_LENGTH( REPLACE(haystack, delim, '')) + 1;
END$$
DELIMITER ;
The VALUE_IN_SET function: treats the delimited list as a one based array and returns the value at the given 'index'.
DELIMITER $$
DROP FUNCTION IF EXISTS `VALUE_IN_SET`$$
CREATE FUNCTION `VALUE_IN_SET`(haystack VARCHAR(1024),
delim CHAR(1),
which INTEGER
) RETURNS VARCHAR(255) CHARSET utf8 COLLATE utf8_unicode_ci
BEGIN
RETURN SUBSTRING_INDEX(SUBSTRING_INDEX(haystack, delim, which),
delim,
-1);
END$$
DELIMITER ;
Related Information:
The tables (with data):
CREATE TABLE `integerseries` (
`id` int(11) unsigned NOT NULL AUTO_INCREMENT,
PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=500 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
/*Data for the table `integerseries` */
insert into `integerseries`(`id`) values (1);
insert into `integerseries`(`id`) values (2);
insert into `integerseries`(`id`) values (3);
insert into `integerseries`(`id`) values (4);
insert into `integerseries`(`id`) values (5);
insert into `integerseries`(`id`) values (6);
insert into `integerseries`(`id`) values (7);
insert into `integerseries`(`id`) values (8);
insert into `integerseries`(`id`) values (9);
insert into `integerseries`(`id`) values (10);
Resource:
CREATE TABLE `resource` (
`id` int(11) NOT NULL,
`data` varchar(250) COLLATE utf8_unicode_ci DEFAULT NULL,
PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
/*Data for the table `resource` */
insert into `resource`(`id`,`data`) values (1,'abcde');
insert into `resource`(`id`,`data`) values (2,'qwerty');
insert into `resource`(`id`,`data`) values (3,'azerty');
User_resource:
CREATE TABLE `user_resource` (
`user` varchar(50) COLLATE utf8_unicode_ci NOT NULL,
`resources` varchar(250) COLLATE utf8_unicode_ci DEFAULT NULL,
PRIMARY KEY (`user`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
/*Data for the table `user_resource` */
insert into `user_resource`(`user`,`resources`) values ('sampleuser','1;2;3');
insert into `user_resource`(`user`,`resources`) values ('stacky','3');
insert into `user_resource`(`user`,`resources`) values ('testuser','1;3');
user table (t1), by using a sub-query, into a 'proper table', then could you solve it in one query? If so, the problem becomes the issue of converting theuser1 | 1;2;4into 'rows'. Is that possible in standard SQL. Yes, if you join it with a table of sequential 'integers'. I call in 'integerseries'. I would put it in a view initially while I tested it. I haven't tried it but it should work.