String to rows

Question

I know this has been asked many times, and I got the simple solution for converting a (comma separated) string to rows like this:
with t as (select 'aa;bb;cc;dd;ee;ff' as txt from dual)
  select regexp_substr (txt, '[^;]+', 1, level)
  from t
  connect by level <= length(regexp_replace(txt,'[^;]*'))+1;This works fine for one string. But my situation is that I have a table like this:
col1    col2
A       aa;bb;cc
B       dd
etc.I want to convert this to:
A  aa
A  bb
A  cc
B  dd
etc.I can do this with a pipelined function or I can do this with a group by or unique in my sql, e.g.:
with t as (select 'A' as what, 'aa;bb;cc' as txt from dual
           union
           select 'B' as what, 'dd' as txt from dual)
  select unique what,REGEXP_SUBSTR (txt, '[^;]+', 1, level)
  from t
  connect by level <= length(regexp_replace(txt,'[^;]*'))+1;
W REGEXP_S
A bb     
A aa     
B dd     
A cc     
4 rows selected.In this example, with only two records A and B, this is no problem. But in the actual table I am using in my WITH clause, this select is performing slowly due to the UNIQUE.
with t as
  (select col1, col2
  from mytable
  select unique col1, regexp_substr (col2, '[^,]+', 1, level)
  from t
  connect by level <= length(regexp_replace(col2,'[^,]*'))+1;Anybody with a suggestion how to improve this statement?

Answer 1

Hi,
It's very confusing to use "CONNECT BY LEVEL <= x" when there's more than one row.
Generate the highest level you'll ever need in a sub-query (called a counter table ) that is based on a one-row table.
Then join to the counter table in your main query.
For example:
WITH     cntr     AS
     SELECT     LEVEL     AS n
     FROM     dual
     CONNECT BY  LEVEL <= 1 + ( SELECT  MAX ( LENGTH ( REGEXP_REPLACE ( col2
                                                                             , '[^;]'
                       FROM    t
SELECT     t.col1
,     REGEXP_SUBSTR ( t.col2
                , '[^;]+'          -- Use + here, not *
                , 1
                , cntr.n
                ) AS col2_substr
FROM     t
JOIN     cntr     ON     cntr.n     <= LENGTH ( REGEXP_REPLACE (col2 || ';', '[^;]'))
ORDER BY  t.col1
,            cntr.n
;Edited by: Frank Kulash on Oct 23, 2009 12:01 PM