Join with dynamic pivot (version 2) - sql

Question

I have some tables with data:
Category
CategoryID CategoryName
1 Home
2 Contact
3 About
Position
PositionID PositionName
1 Main menu
2 Left menu
3 Right menu
...(new row can be added later)
CategoryPosition
CPID CID PID COrder
1 1 1 1
2 1 2 2
3 1 3 3
4 2 1 4
5 2 3 5
How can I make a table like this:
CID CName MainMenu LeftMenu RightMenu
1 Home 1 2 3
2 Contact 4 0 5
3 About 0 0 0
And if a new Category or Position row is added later, the query should reflect the change automatically, e.g:
CID CName MainMenu LeftMenu RightMenu BottomMenu
1 Home 1 2 3 0
2 Contact 4 0 5 0
3 About 0 0 0 0
4 News 0 0 0 0

Answer 1

The following dynamic query seems to work:
declare #columnlist nvarchar(4000)
select #columnlist = IsNull(#columnlist + ', ', '') + '[' + PositionName + ']'
from #Position
declare #query nvarchar(4000)
select #query = '
select *
from (
select CategoryId, CategoryName, PositionName,
IsNull(COrder,0) as COrder
from #Position p
cross join #Category c
left join #CategoryPosition cp
on cp.pid = p.PositionId
and cp.cid = c.CategoryId
) pv
PIVOT (max(COrder) FOR PositionName in (' + #columnlist + ')) as Y
ORDER BY CategoryId, CategoryName
'
exec sp_executesql #query
Some clarification:
The #columnlist contains the dymamic field list, built from the Positions table
The cross join creates a list of all categories and all positions
The left join seeks the corresponding COrder
max() selects the highest COrder per category+position, if there is more than one
PIVOT() turns the various PositionNames into separate columns
P.S. My table names begin with #, because I created them as temporary tables. Remove the # to refer to a permanent table.
P.S.2. If anyone wants to try his hands at this, here is a script to create the tables in this question:
set nocount on
if object_id('tempdb..#Category') is not null drop table #Category
create table #Category (
CategoryId int identity,
CategoryName varchar(50)
)
insert into #Category (CategoryName) values ('Home')
insert into #Category (CategoryName) values ('Contact')
insert into #Category (CategoryName) values ('About')
--insert into #Category (CategoryName) values ('News')
if object_id('tempdb..#Position') is not null drop table #Position
create table #Position (
PositionID int identity,
PositionName varchar(50)
)
insert into #Position (PositionName) values ('Main menu')
insert into #Position (PositionName) values ('Left menu')
insert into #Position (PositionName) values ('Right menu')
--insert into #Position (PositionName) values ('Bottom menu')
if object_id('tempdb..#CategoryPosition') is not null
drop table #CategoryPosition
create table #CategoryPosition (
CPID int identity,
CID int,
PID int,
COrder int
)
insert into #CategoryPosition (CID, PID, COrder) values (1,1,1)
insert into #CategoryPosition (CID, PID, COrder) values (1,2,2)
insert into #CategoryPosition (CID, PID, COrder) values (1,3,3)
insert into #CategoryPosition (CID, PID, COrder) values (2,1,4)
insert into #CategoryPosition (CID, PID, COrder) values (2,3,5)

Answer 2

Since PIVOT requires a static list of columns, I think a dynamic-sql-based approach is really all that you can do: http://www.simple-talk.com/community/blogs/andras/archive/2007/09/14/37265.aspx

Answer 3

As mentioned by several posters, dynamic SQL using the PIVOT command is the way to go. I wrote a stored proc named pivot_query.sql awhile back that has been very handy for this purpose. It works like this:
-- Define a query of the raw data and put it in a variable (no pre-grouping required)
declare #myQuery varchar(MAX);
set #myQuery = '
select
cp.cid,
c.CategoryName,
p.PositionName,
cp.COrder
from
CategoryPosition cp
JOIN Category c
on (c.CategoryId = cp.cid)
JOIN Position p
on (p.PositionId = cp.pid)';
-- Call the proc, passing the query, row fields, pivot column and summary function
exec dbo.pivot_query #myQuery, 'CategoryName', 'PositionName', 'max(COrder) COrder'
The full syntax of the pivot_query call is:
pivot_query '<query>', '<field list for each row>', '<pivot column>', '<aggregate expression list>', '[<results table>]', '[<show query>]'
it is explained more in the comments at the top of the source code.
A couple of advantages of this proc are that you can specify multiple summary functions like max(COrder),min(COrder) etc. and it has the option to store the output in a table in case you want to join the summary data up with other information.

Answer 4

I guess you need to select using PIVOT. By default, pivots only select a static list of columns. There are some solutions on the net dealing with dynamic column pivots, such as here and here.

Answer 5

My suggestion would be to return your data as a simple join and let the front end sort it out. There are some things for which SQL is excellent, but this particular problem seems like something that the front end should be doing. Of course, I can't know that without knowing your full situation, but that's my hunch.