Dynamic field definitions - can this be done in T-SQL? - tsql

Question

I have a requirement that I'm struggling to implement. If possible, I'd like to achieve this with native T-SQL.
I have the following tables:
CUSTOMER
========
ID,
Name
FIELDDEF
========
ID,
Name
FieldType (Char T, N, D for Text, Number or Date)
CUSTOMERFIELD
=============
ID,
CustomerID,
FieldDefID,
CaptureDate,
ValueText,
ValueNumber,
ValueDate
Basically, the purpose of these tables is to provide an extensible custom field system. The idea is that the user creates new field definitions that can be a text, number or date field. Then, they create values for these fields in the ValueText, ValueNumber OR ValueDate field.
Example:
*Customer*
1,BOB
2,JIM
*FieldDef*
1,Mobile,T
1,DateOfBirth,D
*CustomerField*
ID,CustomerID,FieldDefID,CaptureDate,ValueText,ValueNumber,ValueDate
1,1,1,2011-01-1,07123456789,NULL,NULL
2,1,2,2011-01-1,NULL,NULL,09-DEC-1980
3,1,1,2011-01-2,07123498787,NULL,NULL
I need to create a view that looks like this:
*CustomerView*
ID,Name,Mobile,DateOfBirth
1,BOB,07123498787,09-DEC-1980
Note that Bob's mobile is the second one in the list, because it uses the most recent capture date.
Ideally, I need this to be extensible, so if I create a new field def in the future, it is automatically picked up in the CustomerView.
Is this possible in T-SQL at all?
Thanks,
Simon.

Answer 1

This would not be possible with a view, unless the view is dynamically recreated on the fly every time FieldDef changes because view schemas are locked-in at creation time. However, it may be possible with a stored procedure, which may or may not work depending on how you are using it.
Edit 1
Here is a sample query that works just for your current field names, and would have to be modified by dynamic SQL to work in general:
Edit 2
Modified to grab the newest values from the customer field table
with CustomerFieldNewest as (
select
cf1.*
from
customerfield cf1
inner join
(
select
customerid,
fielddefid,
max(capturedate) as maxcapturedate
from
customerfield cf2
group by
customerid,
fielddefid
) cf2 on cf1.customerid = cf2.customerid
and cf1.fielddefid = cf2.fielddefid
and cf1.capturedate = cf2.maxcapturedate
)
,CustomerFieldPivot as (
select
C.ID as ID
,max(case when F.Name = 'Mobile' then CF.ValueText end) as Mobile
,max(case when F.Name = 'DateOfBirth' then CF.ValueDate end) as DateOfBirth
from
Customer C
left join
CustomerFieldNewest CF on C.ID = CF.CustomerID
left join
FieldDef F on F.ID = CF.FieldDefID
group by
C.ID
)
select
C.*
,P.Mobile
,P.DateOfBirth
from
Customer C
left join
CustomerFieldPivot P on C.ID = P.ID
Edit 3
Here is T-SQL code to generate the view on the fly based on the current set of fields in FieldDef (this assumes the view CustomerView already exists, so you will need to create it first as a blank definition or you will get an error). I'm not sure about the performance of all this, but it should work correctly.
declare #sql varchar(max)
declare #fielddef varchar(max)
declare #fieldlist varchar(max)
select
#fielddef = coalesce(#fielddef + ', ' + CHAR(13) + CHAR(10), '') +
' max(case when F.Name = ''' + F.Name + ''' then CF.' +
case F.FieldType
when 'T' then 'ValueText'
when 'N' then 'ValueNumber'
when 'D' then 'ValueDate'
end
+ ' end) as [' + F.Name + ']'
,#fieldlist = coalesce(#fieldlist + ', ' + CHAR(13) + CHAR(10), '') +
' [' + F.Name + ']'
from
FieldDef F
set #sql = '
alter view [CustomerView] as
with CustomerFieldNewest as (
select
cf1.*
from
customerfield cf1
inner join
(
select
customerid,
fielddefid,
max(capturedate) as maxcapturedate
from
customerfield cf2
group by
customerid,
fielddefid
) cf2 on cf1.customerid = cf2.customerid
and cf1.fielddefid = cf2.fielddefid
and cf1.capturedate = cf2.maxcapturedate
)
,CustomerFieldPivot as (
select
C.ID as ID,
' + #fielddef + '
from
Customer C
left join
CustomerFieldNewest CF on C.ID = CF.CustomerID
left join
FieldDef F on F.ID = CF.FieldDefID
group by
C.ID
)
select
C.*,
' + #fieldlist + '
from
Customer C
left join
CustomerFieldPivot P on C.ID = P.ID
'
print #sql
exec(#sql)
select * from CustomerView

Answer 2

You need to build a crosstab which you do with the Pivot statement in TSQL. Here's an article that talks about how to build the pivot dynamically.
http://sqlserver-qa.net/blogs/t-sql/archive/2008/08/27/4809.aspx

Answer 3

Just for completeness there is sql_variant:
declare #t table (typ varchar(1), yuk sql_variant)
insert #t values ('d', getdate())
insert #t values ('i', 1234)
insert #t values ('s', 'bleep bloop')
select
yuk,
case typ
when 'd' then convert(datetime, yuk, 106)+50
when 'i' then cast(yuk as int) * 2
when 's' then reverse(cast(yuk as varchar))
else yuk
end
from #t