Result may contain single string or comma separated need to split into rows

Question

Hi, All... I've searched through the forums and found plenty on splitting comma separated into rows; though I'm struggling applying it to my situation. I have data that looks like the below where I need to split a value into multiple rows if it should be but the same field in the table may also contain a string that should not be separated (indicated by the "Array" field being 0 or 1)...
WITH t AS
(SELECT 1 as array, '"Banana", "Apple", "Pear"' as str FROM dual union all
SELECT 0, 'Fruit is delicious' FROM dual union all
SELECT 0, 'So are vegetables' FROM dual union all
SELECT 1, '"Bean", "Carrot", "Broccoli"' FROM dual union all
SELECT 1, '"Apple", "Banana"' FROM dual)I've looked through many of the connect_by posts on the forum and I haven't come across one that splits a field if it should be but doesn't if it should not be... may have missed it because there are plenty of these requests on the forum!
If you're feeling even more ambitious - the ultimate goal is to count the number of times a particular answer appears in an array - so notice the last portion of the data contains "Apple", "Banana"... the result would show:
RESULT
Banana             2
Apple              2
Pear               1
Bean               1
Carrot             1
Broccoli           1
Fruit is delicious
So are vegetablesI can always sort them later based on other fields in the table - but the result above would be my ultimate goal!
Any help is always appreciated. Thanks! 11g

Answer 1

Hi,
The examples you found should work for you. Just use a CASE expression to determine if str needs to be split or not, by looking at array.
Here's one way:
WITH     got_part_cnt     AS
     SELECT     array, str
     ,     CASE
              WHEN  array = 0
              THEN  1
--              ELSE  1 + REGEXP_COUNT (str, '", "')     -- See note below
              ELSE  1 + ( ( LENGTH (str)
                              - LENGTH (REPLACE (str, '", "'))
                     / 4
          END          AS part_cnt
     FROM    t
,     cntr     AS
     SELECT  LEVEL     AS n
     FROM     (
              SELECT  MAX (part_cnt)     AS max_part_cnt
              FROM    got_part_cnt
     CONNECT BY     LEVEL     <= max_part_cnt
,     got_sub_str          AS
     SELECT     CASE
              WHEN  p.array = 0
              THEN  p.str
              ELSE  REGEXP_SUBSTR ( p.str
                               , '[^"]+'
                         , 1
                         , (2 * c.n) - 1
          END     AS sub_str
     FROM     got_part_cnt  p
     JOIN     cntr           c  ON  c.n <= p.part_cnt
SELECT       sub_str
,       COUNT (*)     AS cnt
FROM       got_sub_str
GROUP BY  sub_str
ORDER BY  cnt          DESC
,            sub_str
{code}
The only database at hand right now is Oracle 10.2, which doesn't have REGEXP_COUNT.  I had to use a complicated way of counting how many times '", "' occurs in str in order to test this in Oracle 10.  Since you have Oracle 11, you can un-comment the line that uses REGEXP_COUNT in the first CASE expression, and remove the alternate ELSE clause (that is, the next 5 lines, up to END).
To make sure that this query is really paying attention to array, I added this row to the sample data:
{code}
SELECT 0, '"Bean", "Carrot", "Broccoli"' FROM dual union all
{code}
Even though str looks just like a delimited list, array=0 tells the query not to split it, so it produces these results:
{code}
SUB_STR                               CNT
Apple                                   2
Banana                                  2
"Bean", "Carrot", "Broccoli"            1
Bean                                    1
Broccoli                                1
Carrot                                  1
Fruit is delicious                      1
Pear                                    1
So are vegetables                       1
{code}