How to show multiple values for Unique records in Report

Question

Here's my question/problem:
I've joined two tables, one table (TBL1) contains an object id (OBJ_ID) that repeats and the other table (TBL2) contains a date (DT), object type id (OBJ_TYP_ID), and object type description (OBJ_TYP_DES). The tables are joined by an inventory id (INV_ID).
The OBJ_ID repeats and has a Date value for each record. I want to report an unique OBJ_ID and show each Date for a particular OBJ_ID in multiple Columns.
An example of the current resultset looks like this:
OBJ_ID OBJ_TYP_ID OBJ_TYP_DES DATE
1 1 TYPE1 4/1/2009
2 1 TYPE1 4/1/2009
3 1 TYPE1 4/10/2009
1 2 TYPE2 5/3/2009
3 1 TYPE1 3/30/2005
4 1 TYPE1 4/1/2009
5 1 TYPE1 4/1/2009
5 2 TYPE2 5/1/2009
1 1 TYPE1 4/3/2007
1 1 TYPE1 3/30/2005
I want to express the resultset like this:
OBJ_ID OBJ_TYP_ID OBJ_TYPE_DES DATE1 DATE2 DATE3
1 1 TYPE1 4/1/2009 4/3/2007 3/30/2005
1 2 TYPE2 5/3/2009
2 1 TYPE1 4/1/2009
3 1 TYPE1 4/10/2009 3/30/2005
4 1 TYPE1 4/1/2009
5 1 TYPE1 4/1/2009
5 2 TYPE2 5/1/2009
What technique is best to use to do this? I know I could create another table and populate the rows/columns by reading data from this query, but is there a better way?

Answer 1

Hi,
cclemmons wrote:
I want to express the resultset like this:
OBJ_ID OBJ_TYP_ID OBJ_TYPE_DES DATE1 DATE2 DATE3
1 1 TYPE1 4/1/2009 4/3/2007 3/30/2005
1 2 TYPE2 5/3/2009
2 1 TYPE1 4/1/2009
3 1 TYPE1 4/10/2009 3/30/2005
4 1 TYPE1 4/1/2009
5 1 TYPE1 4/1/2009
5 2 TYPE2 5/1/2009
What technique is best to use to do this? I know I could create another table and populate the rows/columns by reading data from this query, but is there a better way?Absolutely! You seem to have an instictive feeling that creating a separate table just for a query is inefficient. You instinct is 100% correct. Maybe in Oracle 7 there was a reason to do that, but today you can query results of other queries as if they were tables, so temporary tables like you describe are very rarely necessary, let alone convenient.
What you want to do is pivot the date columns. Here's one way
WITH     original_query     AS
     SELECT  obj_id
     ,     obj_typ_id
     ,     obj_typ_des
     ,     dt          -- date is not a good column name
     FROM     ...          -- the rest of your original query goes here
,     got_rnum     AS
     SELECT     oq.*
     ,     ROW_NUMBER () OVER ( PARTITION BY  obj_id
                               ,                    obj_typ_id
                         ,             obj_typ_des
                         ORDER BY        dt     DESC
                       )         AS rnum
     FROM    original_query        oq
SELECT       obj_id
,       obj_typ_id
,       obj_typ_des
,       MAX (CASE WHEN rnum = 1 THEN dt END)     AS dt1
,       MAX (CASE WHEN rnum = 2 THEN dt END)     AS dt2
,       MAX (CASE WHEN rnum = 3 THEN dt END)     AS dt3
,       MAX (CASE WHEN rnum = 4 THEN dt END)     AS dt4
,       MAX (CASE WHEN rnum = 5 THEN dt END)     AS dt5
FROM       got_rnum
GROUP BY  obj_id
,       obj_typ_id
,       obj_typ_des
;As you can see, this adds two layers of queries on top of your original query. One of those layers is probably not needed; depending on what you're doing in your original main query, you can probably compute rnum there, and omit the got_rnum sub-query.
Also, depending on the relationship of obj_id, obj_typ_id and obj_typ_des, some of what I posted above may not be needed but including it won't really hurt.
If you want to know more about this technique, search for "pivot" or "rows to columns". A <tt>SELECT ... PIVOT ...</tt> keyword was introduced in Oracle 11, but most of what you'll find when you search for "pivot" doesn't assume you have Oracle 11 (nor does the query above require Oracle 11).
This assumes you know an upper limit (5 in the example above) of dts that can appear in any line of output.
See [this thread|http://forums.oracle.com/forums/thread.jspa?messageID=3527823&#3527823] for a discussion of some alternatives.