How to do pivoting on part of an SQL output row in 10g

Question

Hi,
I'm using Oracle 10.1.0.5.0.
I would like to know what the general decode is for pivoting in 10g. I need to display PAGE_DISPLAY_NAME, ITEM_DISPLAY_NAME, ITEM_TYPE_DISPLAY_NAME essentially once and then display ATTRIBUTE_DISPLAY_NAME and ITEM_ATTRIBUTE_VALUE as a separate column for each:
So my query starts me with the following output.
PAGE_DISPLAY_NAME        ITEM_DISPLAY_NAME            ITEM_TYPE_DISPLAY_NAME     ATTRIBUTE_DISPLAY_NAME                              ITEM_ATTRIBUTE_VALUE
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Benefit                                                       0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Accounting                               0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Consulting Services                       0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Customer Data Management              0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Facilities                                0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Finance                                0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: HR                                        0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: IT                                        0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: International                        0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Inventory Control                     0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Legal                                0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Marketing                                0
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Sales                                1
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record     Business Function: Sales Operations                     1I'd like to show the following:
PAGE_DISPLAY_NAME        ITEM_DISPLAY_NAME            ITEM_TYPE_DISPLAY_NAME        decode(...),   decode(...), .....         decode(etc...
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record          Benefit    Business Function: Accounting  
Content     Rpt-Yearly-EMC Transactions-2007     RIKR Content Record             0                   0                              What's the general SQL for doing this type of thing?
For the record, here's the initial SQL:
select p.DISPLAY_NAME               PAGE_DISPLAY_NAME,
i.DISPLAY_NAME                      ITEM_DISPLAY_NAME,
it.DISPLAY_NAME                     ITEM_TYPE_DISPLAY_NAME,
attr.DISPLAY_NAME                   ATTRIBUTE_DISPLAY_NAME,
ia.VALUE                            ITEM_ATTRIBUTE_VALUE
from
wwsbr_item_attributes ia,
wwsbr_attributes attr,
wwsbr_all_items i,
wwsbr_item_types it,
wwsbr_all_folders p
where
    ia.attribute_id = attr.id
and ia.attribute_caid = attr.CAID
and ia.item_masterid = i.id
and ia.item_caid = i.caid
and i.subtype = it.id
and i.subtype_caid = it.caid
and i.folder_id = p.id
and i.active = 1
and i.visible = 1
and i.is_current_version = 1
and upper(it.DISPLAY_NAME) like '%RIKR%'
and substr(attr.DISPLAY_NAME,1,5) <> '-----' 
order by i.DISPLAY_NAME, attr.DISPLAY_NAME;

Answer 1

Hi,
There are a few errors in the INSERT statements, but it looks like the ones that work provide a good enough sample.
In your sample data,
attribute_display_name = 'Benefit' only when item_type_display_name LIKE '%RIKR%', and
attribute_display_name = 'Billing & Collections' only when item_type_display_name LIKE '%OMKR%'
and so on, for the other values. In other words, just by looking at attribute_display_name , once could predict whether item_type_display_name was LIKE '%RIKR%' or '%OMKR%'.
If that's alwsays the case, then you can say:
SELECT    page_display_name
,       item_display_name
,       item_type_display_name
,       MAX (DECODE ( attribute_display_name
                      , 'Benefit'                         , item_attribute_value
                      , 'Billing & Collections'                    , item_attribute_value
           )          AS p1
,       MAX (DECODE ( attribute_display_name
                       , 'Business Function: Accounting'               , item_attribute_value
                , 'Change Description'                    , item_attribute_value
           )           AS p2
,       MAX (DECODE ( attribute_display_name
                      , 'Business Function: Consulting Services'     , item_attribute_value
                , 'Deal Type: Counrty Federal'                 , item_attribute_value
           )           AS p3
,       MAX (DECODE ( attribute_display_name
                , 'Deal Type: Counrty International Order'     , item_attribute_value
           )           AS p4
FROM       taxonomy
GROUP BY  page_display_name
,       item_display_name
,       item_type_display_name
ORDER BY  page_display_name
,       item_display_name
,       item_type_display_name
;This produces generically-named columns, 'p1', 'p2', ...:
                           ITEM_
PAGE_      ITEM_           TYPE_
DISPLAY_   DISPLAY_        DISPLAY_
NAME       NAME            NAME                 P1  P2  P3  P4
Archive    Rpt-AR          RIKR Content Record      1   0
Content    Rpt-Booking     RIKR Content Record      1   0
Documents  Calc-Discount   OMKR Document        0       0   0
Documents  Calc-Early      OMKR Document        0       0   0I shortened item_display_name to make the output more readable.
If you add a WHERE clause like
WHERE   item_type_display_name LIKE '%RIKR%'then, of course, you would only get the first two rows of output from the result set above, but you would get all the same columns, including the p4 column that will necessarily be NULL.
If my earlier assumption was wrong (for example, if attribute_display_name can 'Benefit' even when item_type_display_name is NOT LIKE '%RIKR%', but, when that happens, you want to ignore it), then change the pivot columns like this:
,       MAX ( CASE
                     WHEN  (     item_type_display_name LIKE '%RIKR%'
                       AND  attribute_display_name = 'Benefit'
               OR        (     item_type_display_name LIKE '%OMKR%'
                       AND  attribute_display_name = 'Billing & Collections'
               THEN  item_attribute_value
          END
           )          AS p1
,       MAX ( CASE
                     WHEN  (     item_type_display_name LIKE '%RIKR%'
                       AND  attribute_display_name = 'Business Function: Accounting'
               OR        (     item_type_display_name LIKE '%OMKR%'
                       AND  attribute_display_name = 'Change Description'
               THEN  item_attribute_value
          END
           )          AS p2
,       MAX ( CASE
                     WHEN  (     item_type_display_name LIKE '%RIKR%'
                       AND  attribute_display_name = 'Business Function: Consulting Services'
               OR        (     item_type_display_name LIKE '%OMKR%'
                       AND  attribute_display_name = 'Deal Type: Counrty Federal'
               THEN  item_attribute_value
          END
           )          AS p3
,       MAX ( CASE
                     WHEN  (     item_type_display_name LIKE '%OMKR%'
                       AND  attribute_display_name = 'Deal Type: Counrty International Order'
               THEN  item_attribute_value
          END
           )          AS p4For the sample data given, this produces the same output as above.
All this assumes that the conditions
item_type_display_name LIKE '%RIKR%' and
item_type_display_name LIKE '%OMKR%' are mutualyy exclusive; that is, you never have item_type_display like 'RIKR Content Record/OMKR Document'.
If you do have values like that, then you'll need separate columns for each of the possible values of attribute_display_name. Given your sample data, that means 7 pivoted columns. Depending on your WHERE clause and your data, some of those columns might by NULL for all rows. As you suggested, you coul;d use the anlaytic ROW_NUMBER function to number the rows that actually occurred in the query. You would still have 7 pivoted columns, but the ones that were not always NULL would appear on the left, where they were easier to read.
The number of columns, and their aliases, must be hard-coded into the query. If you want something that has column aliases like BENEFIT instead of P1, or only has 4 piovted columns when you only need 4, then you have to use dynamic SQL.
You can fake column headings by doing a separate query, which includes some of the same conditions as the main query, figures out what the pivoted columns will be, and displays appropriate headings. I've done this when producing csv files, where the heading only had to appear once, at the very beginning. Getting such a heading to appear after, say, every 50 rows of output is much more complicated.
You can fake the number of columns by using string aggregation to put all the pivoted data into one humongeous VARCHAR2 column, concatenating spaces so that it looks like 3, or 4, or however many columns.