MySQL store array which can be crossreferenced

Question

I have a MySQL database with 3 tables,

computers (which has the following columns):

      computersid (Unique Key)
      Name
      locationid
      softwareid

locations

      locationid
      location
      campus
      buildingfloor
      room

software

      softwareid
      softwarename
      softwareversion

At the moment, I can easily work out what computers are in what location, because the locationid in computers is a single digit which corresponds to the locationid in the locations table.

What I'm struggling with is working out as a summary, what location has what software. I've been able to make a query that can find all computers with the same locationid (so all computers in a room), but I'm finding it difficult to count the same instances of software a location has, because the 'softwareid' column in the computers table is stored as a varchar array, using "#" as a deliminator. For example:

locationid 5 has the following computers with software id's:

computername: TEST1 softwareid: #1# #4# #53# #67# #32#

computername: TEST2 softwareid: #7# #1# #55# #36# #92#

computername: TEST3 softwareid: #2# 67# #7# #5# #2#

I'm struggling to think of an efficient method to sift through this data and count how many instances of software a location has, and am questioning whether my method for storing the array is correct what what I'm trying to achieve.

The output I'm aiming for, is a table that has all the distinct software in a location, along with how many computers has that software installed.

Any help or tips would be greatly appreciated; I'm new to SQL.

Answer 1

Oh, my my my, what a mess of a database schema! This business of creating a one-to-many relationship by putting a list of values in a string makes life hard. What you really need is a separate computer_software join table.

You have a saving grace here. You have a list of all the possible software items in their own table, and your string format is nicely delimited.

So, you can issue this SQL query (http://sqlfiddle.com/#!2/36a672/4/0) to JOIN together the computers and software tables.

SELECT c.name, c.id, s.name AS sname,s.version
  FROM software AS s
  JOIN computers AS c ON c.softwareid LIKE CONCAT('%#',s.id,'#%')
 ORDER BY c.name, s.name, s.version

This will perform horribly. But, so what, it will still be faster than starting over with your design. It uses a whole lot of operations like this to match up the two tables.

    '#1# #4# #53# #67# #32#' LIKE  '%#53#%'

It uses the CONCAT() function to create strings such as '%#53#%'

So, I suggest you move forward by creating a view computers_software, with this command.

CREATE VIEW computers_software AS
    SELECT c.id AS computer_id, s.id AS software_id
    FROM software AS s
    JOIN computers AS c ON c.softwareid LIKE CONCAT('%#',s.id,'#%')

This view -- this virtual table -- this virtual join table -- contains pairs of (computer_id, software_id) values. The presence of a row means that the specified computer has the specified software.

Then to determine what software is in what location, you do this (not debugged):

SELECT L.location, s.name, s.version, COUNT(*) AS copies
  FROM location AS L
  JOIN computers AS c ON l.id = c.locationid
  JOIN computers_software AS cs ON c.id = cs.computer_id
  JOIN software AS s ON cs.software_id = s.id
 GROUP BY  L.location, s.name, s.version

Again, the trick to doing this in a sql-fluent way is to build the many-to-many relationship called computers_software. I'm suggesting you do it as a view for now.

Ordinarily an application developer would create a table like that simply because it works better. You might consider making that switch.