Querying all_tab_columns

Question

Hi all,
This is a "my query takes too long" question but it relates to the ALL_TAB_COLUMNS view.
I'm trying to generate some basic DML scripts from a large number of tables (insert/update) the cheating way by selecting columns from all_tab_columns
e.g.
select 'insert into '||table_name||' ('
  from all_tables
where table_name = 'TABLE_NAME_HERE'
union all
select column_name||','
  from all_tab_columns
where table_name = 'TABLE_NAME_HERE'
--etc.however, I want the column names all on one line.
so I have the following, which I imagine should create the column_name's on one line but it "runs for ages"
select sys_connect_by_path(column_name,', ') c
  from all_tab_columns
where owner = 'WHITEHAT'
   and table_name = 'FOO'
   and connect_by_isleaf = 1
   start with column_id = 1
   connect by column_id = prior column_id + 1I am using 10gR2 so unfortunately I can't use recursive subquery factoring just yet, although that equivalent runs quickly as expected on our 11g environment:
with t (column_name, column_id)
             as (select cast(column_name as varchar2(4000)), column_id
                   from all_tab_columns
                  where owner = 'WHITEHAT'
                    and table_name = 'FOO'
                    and column_id = 1
                  UNION ALL
                  SELECT cast(t.column_name||', '||atc.column_name as varchar2(4000))
                        ,atc.column_id
                    from t
                        ,all_tab_columns atc
                   where owner = 'WHITEHAT'
                    and table_name = 'FOO'
                    and atc.column_id = t.column_id + 1
select column_name
  from (
         select column_name, column_id, max(column_id) over () max_col_id
         from t          
where column_id = max_col_idwhat am I doing wrong?

Answer 1

Hi,
WhiteHat wrote:
... so I have the following, which I imagine should create the column_name's on one line but it "runs for ages"
select sys_connect_by_path(column_name,', ') c
from all_tab_columns
where owner = 'WHITEHAT'
and table_name = 'FOO'
and connect_by_isleaf = 1
start with column_id = 1
connect by column_id = prior column_id + 1... what am I doing wrong?The WHERE clause is applied after the START WITH and CONNECT BY clauses are complete.
So you're starting with one row from every table. Let's say you have a small database, with 1000 tables. On LEVEL=1 you have 1000 rows. Then you're connecting every column with column_id=2 to each of those those 1000 rows. Almost every table will have a column with column_id=2, so you'll have 1,000,000 rows on LEVEL=2. I think you can see where this is going.
Don't use a WHERE clause to get the right table. Put the conditions you now have in the WHERE clause (or equivalent conditions, except for "connect_by_isleaf = 1") in the START WITH and CONNECT BY clauses, instead, or apply the WHERE clause in a sub-query, first, like this:
WITH     universe     AS
     SELECT     column_name
     ,     column_id
     FROM     all_tab_columns
     WHERE     owner          = 'WHITEHAT'
     AND     table_name     = 'FOO'
SELECT     SYS_CONNECT_BY_PATH (column_name, ',')     AS c
FROM     universe
WHERE     CONNECT_BY_ISLEAF     = 1
START WITH     column_id     = 1
CONNECT BY     column_id     = 1 + PRIOR column_id
;