Pivoting Multiple Columns Based On a Single Column

Question

I tried following several other suggestions on this website but coulnd't make this work for some reason.

Basically I'm trying to get from this table of Tasks:

| TaskID | ProjectID | TaskType | TaskDate1 | TaskDate2 |
---------------------------------------------------------
|      1 |         1 |    Type4 |  20130401 |  20130506 |
|      2 |         1 |    Type0 |  20130412 |  20130508 |
|      3 |         1 |    Type2 |  20130420 |  20130517 |

To that one:

| ProjectID | Type0Date1 | Type0Date2 | Type2Date1 | Type2Date2 | Type4Date1 | Type4Date2 |
---------------------------------------------------------
|         1 |   20130412 |   20130508 |   20130420 |   20130517 |   20130401 |   20130506 |

Each project has to have each of several predefined tasks (identified by their type). These tasks have a lot of properties but I'm interested in 2 dates and it would be nice to bring them all in one row per project.

I can easily pivot it for one of the columns with: PIVOT (MAX(TaskDate1) FOR TaskType IN ([Type0],[Type2],[Type4]))

I tried duplicating the TaskType column and doing another pivot around it for TaskDate2 but that won't give me a single row.

I also can't wrap my head around how unpivoting the table first would get me where I need to be, as was the suggestion in some other posts for similar questions.

If there is a scalable way of doing this (as in pivoting even more columns) it would be nice to know.

Thanks.

Answer 1

You can get the desired result but since you need to PIVOT on multiple columns ( TaskDate1 and TaskDate2 ) I would first unpivot the two columns into multiple rows.

Since you are using SQL Server 2008+, you can use either the UNPIVOT function or CROSS APPLY to convert the multiple columns. The basic syntax will be:

select projectid,
  col = TaskType+col,
  value
from yourtable
cross apply
(
  values 
    ('Date1', TaskDate1),
    ('Date2', TaskDate2)
) c(col, value)

See SQL Fiddle with Demo . This will convert your data into an easy to use format:

| PROJECTID |        COL |                        VALUE |
|-----------|------------|------------------------------|
|         1 | Type4Date1 | April, 01 2013 00:00:00+0000 |
|         1 | Type4Date2 |   May, 06 2013 00:00:00+0000 |
|         1 | Type0Date1 | April, 12 2013 00:00:00+0000 |
|         1 | Type0Date2 |   May, 08 2013 00:00:00+0000 |

Once you have the data in this format, then you can apply the PIVOT function to the get the max(value) for each col :

select projectId,
  Type0Date1, Type0Date2, 
  Type2Date1, Type2Date2,
  Type4Date1, Type4Date2
from
(
  select projectid,
    col = TaskType+col,
    value
  from yourtable
  cross apply
  (
    values 
      ('Date1', TaskDate1),
      ('Date2', TaskDate2)
  ) c(col, value)
) d
pivot
(
  max(value)
  for col in (Type0Date1, Type0Date2, Type2Date1, Type2Date2,
              Type4Date1, Type4Date2)
) piv;

See SQL Fiddle with Demo .

The above version will work great if you have a limited number of values per ProjectId , but if you have an unknown number of Tasks for each project, then you will need to use dynamic SQL to get the final result:

DECLARE @cols AS NVARCHAR(MAX),
    @query  AS NVARCHAR(MAX)

select @cols = STUFF((SELECT ',' + QUOTENAME(TaskType+col) 
                    from yourtable
                    cross apply
                    (
                      select 'Date1', 1 union all
                      select 'Date2', 2
                    ) c(col, so)
                    group by col, tasktype, so
                    order by tasktype, so
            FOR XML PATH(''), TYPE
            ).value('.', 'NVARCHAR(MAX)') 
        ,1,1,'')

set @query = 'SELECT projectid, ' + @cols + ' 
            from 
            (
              select projectid,
                col = TaskType+col,
                value
              from yourtable
              cross apply
              (
                values 
                  (''Date1'', TaskDate1),
                  (''Date2'', TaskDate2)
              ) c(col, value)
            ) x
            pivot 
            (
                max(value)
                for col in (' + @cols + ')
            ) p '

execute sp_executesql @query

See SQL Fiddle with Demo . These will give a result:

| PROJECTID |                   TYPE0DATE1 |                 TYPE0DATE2 |                   TYPE2DATE1 |                 TYPE2DATE2 |                   TYPE4DATE1 |                 TYPE4DATE2 |
|-----------|------------------------------|----------------------------|------------------------------|----------------------------|------------------------------|----------------------------|
|         1 | April, 12 2013 00:00:00+0000 | May, 08 2013 00:00:00+0000 | April, 20 2013 00:00:00+0000 | May, 17 2013 00:00:00+0000 | April, 01 2013 00:00:00+0000 | May, 06 2013 00:00:00+0000 |