[10g] Need help with order by clause in hierarchical query

Question

I have the following sample data:
CREATE TABLE     bill_test1
(     parent_part     CHAR(25)
,     child_part     CHAR(25)
,     line_nbr     NUMBER(5)
,     qty_per          NUMBER(9,5)
INSERT INTO bill_test1 VALUES ('ABC-1','ABC-10',100,1);
INSERT INTO bill_test1 VALUES ('ABC-1','ABC-20',200,2);
INSERT INTO bill_test1 VALUES ('ABC-1','ABC-30',300,3);
INSERT INTO bill_test1 VALUES ('ABC-1','HARDWARE-1',401,10);
INSERT INTO bill_test1 VALUES ('ABC-1','HARDWARE-2',402,5);
INSERT INTO bill_test1 VALUES ('ABC-10','ABC-155',100,2);
INSERT INTO bill_test1 VALUES ('ABC-10','HARDWARE-1',200,1);
INSERT INTO bill_test1 VALUES ('ABC-155','RAW-2',100,4.8);
INSERT INTO bill_test1 VALUES ('ABC-155','HARDWARE-3',200,3);
INSERT INTO bill_test1 VALUES ('ABC-20','RAW-1',100,10.2);
INSERT INTO bill_test1 VALUES ('ABC-30','RAW-3',100,3);And the query below gives me exactly what I want, in the order I want it. However, I am wondering if there is a way to get this order without creating the SEQ column, since I don't need it in my results
SELECT     part_nbr
,     parent_part
,     child_part
FROM     (
     SELECT     CONNECT_BY_ROOT b.parent_part                         AS part_nbr
     ,     b.parent_part
     ,     b.child_part
     ,     SYS_CONNECT_BY_PATH(b.line_nbr,' ')                    AS seq
     FROM     bill_test1 b
     ,     dual
     CONNECT BY     parent_part     = PRIOR child_part
WHERE          part_nbr     = 'ABC-1'
ORDER BY     seq
Results of above query, except with SEQ included in SELECT (just to show what I'm sorting off of):
PART_NBR                     PARENT_PART                  CHILD_PART                   SEQ
ABC-1                        ABC-1                        ABC-10                        100
ABC-1                        ABC-10                       ABC-155                       100 100
ABC-1                        ABC-155                      RAW-2                         100 100 100
ABC-1                        ABC-155                      HARDWARE-3                    100 100 200
ABC-1                        ABC-10                       HARDWARE-1                    100 200
ABC-1                        ABC-1                        ABC-20                        200
ABC-1                        ABC-20                       RAW-1                         200 100
ABC-1                        ABC-1                        ABC-30                        300
ABC-1                        ABC-30                       RAW-3                         300 100
ABC-1                        ABC-1                        HARDWARE-1                    401
ABC-1                        ABC-1                        HARDWARE-2                    402

Answer 1

Hi,
As long as there's only one root, you can say ORDER SIBLINGS BY, but you can't do that in a sub-query (well, you can, but usually there's no point in doing it in a sub-query). If the CONNECT BY is being done in a sub-query, there is no guarantee that the main query will preserve the hierarchical order that the sub-query provides.
The query you posted doesn't require a suib-query, so you can say:
SELECT     CONNECT_BY_ROOT b.parent_part                         AS part_nbr
,     b.parent_part
,     b.child_part
--,     SYS_CONNECT_BY_PATH(b.line_nbr,' ')                    AS seq
FROM     bill_test1 b
WHERE          CONNECT_BY_ROOT b.parent_part     = 'ABC-1'
CONNECT BY     parent_part     = PRIOR child_part
ORDER SIBLINGS BY     b.line_nbr     
;I said the query you posted doesn't require a sub-query. It also doesn't require dual, so I suspect what you posted is a simplification of what you're really doing, and that may need a sub-query. In particular, if you intend to GROUP BY part_nbr, then you need the sub-query. We can repeat the CONNECT_BY_ROOT expression in the WHERE clause (or, now that I think about it, use a START WITH clause instead of WHERE), but, for some reason, we can't use CONNECT_BY_ROOT in a GROUP BY clause; we need to compute CONNECT_BY_ROOT in a sub-query, give it a name (like part_nbr), and GROUP BY that column in a super-query.
This assumes that there is only one root node. ORDER SIBLINGS BY means just that: children of a common parent will appear in order, but the root nodes, who have no parents, will not necessarily be in order.
Here's what I meant by using START WITH instead of WHERE:
SELECT     CONNECT_BY_ROOT b.parent_part                         AS part_nbr
,     b.parent_part
,     b.child_part
--,     SYS_CONNECT_BY_PATH(b.line_nbr,' ')                    AS seq
FROM     bill_test1 b
START WITH     b.parent_part     = 'ABC-1'
CONNECT BY     parent_part     = PRIOR child_part
ORDER SIBLINGS BY     b.line_nbr     
;This should be much more efficient, because it narrows down the results before you waste time getting their descendants.
Using a START WITH clause here is analagous to me sending you an e-mail, saying "Come to a meeting a my office at 3:00."
Using a WHERE clause here is analagous to me sending an e-mail to everyone in the company, saying "Come to a meeting a my office at 3:00", and then, as people get here, telling everyone except you that they can go back.
ORDER SIBLINGS BY was introduced in Oracle 9.
Edited by: Frank Kulash on Dec 9, 2010 2:39 PM
Added version with START WITH clause