Hierachical statement with regular expression

Question

Hi guys,
I was wondering if anyone can help me with the following problem.
Imagine the following tables :
CREATE TABLE organisations
  organ_id number(10),
  parent_id number(10),
  organ_name varchar2(255)
create table renamings
  organ_id number(10),
  search_pattern varchar2(255),
  replace_pattern varchar2(255),
  with_descendants char(1)
INSERT INTO organisations VALUES(0, null, 'Top level');
INSERT INTO organisations VALUES(1, 0, 'A');
INSERT INTO organisations VALUES(11, 1, 'Allo');
INSERT INTO organisations VALUES(2, 0, 'B');
INSERT INTO organisations VALUES(20, 2, 'Battle');
INSERT INTO organisations VALUES(22, 2, 'Star');
INSERT INTO organisations VALUES(299, 2, 'Better');
INSERT INTO organisations VALUES(120, 2, 'Bueno');
INSERT INTO organisations VALUES(3, 0, 'C');
INSERT INTO renamings VALUES(1, 'A', 'X', '0'); -- rename A by X
INSERT INTO renamings VALUES(2, '^B', 'Z', '1'); -- if it starts with B, replace by Z with the descendants.The table organisations contains a hierarchy like an organisational tree. On the renamings table, I have records about the organisations that has to be renamed.
For example, the first record that I inserted in the renamings table means: "Rename the organisation with id = 1. The search patter to apply is 'A' and the replace patter is 'X'. I'd like to use the regexp_replace function to apply the renaming. The latest attribute (0) means that I don't want to rename descendants.
The second record means, rename the organisation with id = 2. The search pattern is '^B' (start with B) and the replace pattern is 'Z'. The last attribute (1) means that the same renaming should be performed on all the descending nodes starting from the organ_id = 2.
The result I would like is:
{code}
ORGAN_ID PARENT_ID ORGAN_NAME
0 Top level
1 0 X
11 1 Allo
2 0 Z
20 2 Zattle
22 2 Star
299 2 Zetter
120 2 Zueno
3 0 C
{code}
I know how to construct the tree, how to rename a specific node using regexp_replace, like this:
{code}
SELECT a.organ_id, a.organ_name, b.search_pattern,
regexp_replace(a.organ_name, b.search_pattern, b.replace_pattern)
FROM organisations a
LEFT JOIN renamings b ON a.organ_id = b.organ_id
START WITH a.organ_id = 0
CONNECT BY PRIOR a.organ_id = a.parent_id;
{code}
But I don't see how I can apply it on all the descending nodes. Any idea?
Thanks,

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statements; that's very helpful.
Always way which version of Oracle you're using. This is especially important with CONNECT BY queries, because every version since Oracle 7 has had significant changes in this area. Since you want to use REGEXP_REPLACE, I'll assume you're using Oracle 10.1 (or higher).
For each row in organisations, you need to know if a row in renamings matches that row or any of its ancestors. We can use SYS_CONNECT_BY_PATH to get a string containing all of the ancestors' organ_ids, ending with the organ_id of the row itself, and use that to find matching rows in renamings:
WITH     got_ancestor_list     AS
     SELECT     organ_id, parent_id, organ_name
     ,     ROWNUM                              AS rnum
     ,     SYS_CONNECT_BY_PATH (organ_id, ',') || ','     AS ancestor_list
     FROM     organisations
     START WITH     organ_id     = 0
     CONNECT BY     parent_id     = PRIOR organ_id
SELECT       a.organ_id, a.parent_id
,       REGEXP_REPLACE ( a.organ_name
                    , r.search_pattern
                , r.replace_pattern
                )          AS new_organ_name
FROM             got_ancestor_list  a
LEFT OUTER JOIN      renamings         r  ON  a.ancestor_list  LIKE '%,'
                                         ||  r.organ_id
                                         || ','
                                         || CASE
                                                                     WHEN  r.with_descendants = '1'
                                              THEN  '%'
                                                                 END
ORDER BY  a.rnum
;For renamings where with_descendants='0', the join condition evaluates to something like:
ON  a.ancestor_list  LIKE '%,1,'which means the last item in ancestor_list must be 1, and the last item in ancestor_lsit is the organ_id of the row itself.
For renamings where with_descendants='1', the join condition evaluates to something like:
ON  a.ancestor_list  LIKE '%,2,%'(with a '%' at the end) which means some item in ancestor_list must be 2, so the condition applies to organ_id=2 and to all its descendants.
This solution assumes that each row in organisations will match (at most) only one row in renamings. If not, which row from renamings do you want to apply? If all of them, in what order? If this is possible in your application, then post examples in your sample data, results and explanation.
This also assumes that renamings.organ_id is in the tree formed by your START WITH conditions. For example, if you changed the START WITH clause to
START WITH     organ_id     = 120then the row with organ_id=120 would keep the organ_name='Bueno', and not be changed to 'Zueno'. Why? Because, if you start with 120, then 120 has no ancestors; it is not a descendant of any other row, such as the row with organ_id=2, and so the renamings on descendants of organ_id=2 wouldn't apply. If your START WITH condition is
START WITH     parent_id     IS NULLthen there is no problem; all relevant renamings will always apply.
By the way, the CHAR data type is nothing but trouble. Use VARCHAR2 instead. I can hear you saying "The maximum length is 1, so it doesn't matter if iwith_descnedants is CHAR or VARCHAR2." If it doesn't matter, then you shouldn't have any objection to using VARCHAR2. However, if the only possible values are '0' and '1', why not make it NUMBER (1)?