Hierarchy query with dynamic start condition

Question

Hi,
I have a table that stores hierarchical items (foohierarchy) and another one storing users and such items (useritems). Unfortunately the latter one is not consistent and misses some entries. I would like to fix that using pure sql. Fixing here means that a user that possesses a certain child element should also have parents of that child. There are a couple of obstacles here though. First the hierarchy table does not explicitly state which elements are top level elements. I am trying to fix that by emulating a proper structure and named that fixedhierarchy. A select statement that returns all intended user/item relationships would be nice.
Here some visualization and sample data, the sample data follows the pattern that each item uses its level as its suffix, but this shouldn't be exploited in the statement:
asci art:
-- A1 (peter,paul)
--   \_ B2
--   \_ C2
--        \_D3 (thomas)
--            \_X4
-- E1
--   \_F2
-- G1
--   \_H2 (heidi)sample data
drop table foohierarchy;
drop table useritems;
create table foohierarchy (
child_item varchar2(30),
parent_item varchar2(30));
insert into foohierarchy values ('B2','A1');
insert into foohierarchy values ('C2','A1');
insert into foohierarchy values ('D3','C2');
insert into foohierarchy values ('X4','D3');
insert into foohierarchy values ('F2','E1');
insert into foohierarchy values ('H2','G1');
create table useritems (
username varchar2 (40),
item varchar2(30)
insert into useritems values ('peter','A1');
insert into useritems values ('paul','A1');
insert into useritems values ('thomas','D3');
insert into useritems values ('heidi','H2');
commit;my attempt at fixing the hierarchy table
with fixedhierarchy as
(select parent_item,child_item from foohierarchy
union all
select distinct null parent_item,parent_item child_item from foohierarchy where parent_item not in (select child_item from foohierarchy))
select * from fixedhierarchy;What the select should return:
'peter'|'A1'
'paul'|'A1'
'thomas'|'A1'
'thomas'|'C2'
'thomas'|'D3'
'heidi'|'G1'
'heidi'|'H2'

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statments; that's very helpful! Don't forget to mention what version of Oracle you're using. That's especially important with hierachies; every version since Oracle 7 has had significant improvements in how to handle them.
I'm a little comfused as to what you want. Are you trying to permanently change the tables? If so, show what the changed table(s) should look like when the job is finished. Post the output you would want to see from "SELECT * FROM useritems;" or "SELECT * FROM foohierarchy;", or, if you want to change both tables, from both.
It looks like you have two separate problems:
(1) Changing foohierarchy, so that every item appears once as a child_id. That means adding rows with NULL parent_item for all the roots ('A1', 'E1' and 'G1'). The query you posted seems directed toward this.
(2) Denormalizing useritems, so that when a username is linked to an item, the username is linked to all that item's ancestors, too. The results you posted seem to be what you want useritems to look like after this is done.
Here's how to do those things:
<h4>(1) Changing foohierarchy</h4>
The query you posted looks good. Do you see anything wrong with it? You can use that query (or something very close to it) in an INSERT or MERGE statement.
If you just want to add some new rows to foohierarchy, then you don't need the UNION: just the second branch of it. If you're cvertain that none of those rows already exist in foohierarchy, then
INSERT INTO foohierarchy (parent_item,      child_item)
     SELECT DISTINCT   NULL,          childitem
     FROM     foohierarchy
     WHERE     parent_item     NOT IN (
                         SELECT      child_item
                         FROM     foohierarchy
;If some of the roots might already be in foohierarchy correctly, then you'll want a something in the WHERE clause to make sure you don';t add a duplicate. A MERGE statment might do this better than INSERT.
<h4>(2) Denormalizing useritems</h4>
This adds new rows to useritems, to make it look like what you posted:
ERGE INTO     useritems     dst
USING (
     WITH     all_ancestors     AS
          SELECT     CONNECT_BY_ROOT child_item     AS descendant
          ,     parent_item               AS ancestor
          FROM     foohierarchy
          START WITH     child_item   IN     (
                                   SELECT     item
                                   FROM     useritems
          CONNECT BY     child_item   = PRIOR parent_item
     SELECT DISTINCT
          u.username
     ,     a.ancestor
     FROM     all_ancestors     a
     JOIN     useritems     u  ON     a.descendant     = u.item
      )                    src
ON     (     src.username     = dst.username
     AND     src.ancestor     = dst.item
WHEN NOT MATCHED THEN
     INSERT     (dst.username,     dst.item)
     VALUES     (src.username,     src.ancestor)
;After doing this, "SELECT * FROM useritems;" will produce this output:
USERNAME   ITEM
paul       A1
peter      A1
thomas     A1
thomas     C2
thomas     D3
heidi      G1
heidi      H2Here's how it works:
Sub-query all_ancestors is a bottom-up CONNECT BY query, linking each node in useritems with all of its ancestors. CONNECT BY queries are often slow under the best circumstances; doing a join in a CONNECT BY query makes it even slower. It's more efficient to do the CONNECT BY query first, in a sub-query, and then join tht result set to other tables, as I did in src.