How does nocycle in connect by work?

Question

hello
there is a table t that defines parent-child relation:
p | c |
6 | 4 |
5 | 4 |
4 | 5 |
4 | 6 |with this query:
select level l, p, c, connect_by_iscycle cycle, connect_by_root p root_p, connect_by_root c root_c
from t connect by nocycle prior c = p;the result for level 3 is the following:
L | P | C | CYCLE | ROOT_P | ROOT_C
3 | 4 | 5 | 1     | 4      | 6  
3 | 4 | 6 | 1     | 4      | 5   my reasoning is as follows. there are 4 rows at level 1: (4,5) (4,6) (5,4) (6,4) and each row can be a starting point for CONNECT BY paths down to level 3
(a) (4,5) -> (5,4) -> (4,5)
(b) (4,5) -> (5,4) -> (4,6)
(c) (4,6) -> (6,4) -> (4,5)
(d) (4,6) -> (6,4) -> (4,6)
(e) (5,4) -> (4,5) -> (5,4)
(f) (5,4) -> (4,6) -> (6,4)
(g) (6,4) -> (4,5) -> (5,4)
(h) (6,4) -> (4,6) -> (6,4)for me (a),(d),(e),(h) are loops, which leaves (b),(c),(f) and (g) as non loops.
my question is: why are (f) and (g) excluded from the result? I suppose I make a mistake in reasoning, but exactly at which point?
thank you
create table t (p number, c number);
insert into t values(4,5);
insert into t values(4,6);
insert into t values(5,4);
insert into t values(6,4);

Answer 1

Hi,
As always, start with the Oracle documentation. The SQL language manual
http://docs.oracle.com/cd/B28359_01/server.111/b28286/pseudocolumns001.htm#i1009313
says "The CONNECT_BY_ISCYCLE pseudocolumn returns 1 if the current row has a child which is also its ancestor. Otherwise it returns 0."
In this example, CONNECT_BY_ISCYCLE=1 when the current row (let's call it X) is connected to another row (Y) according to the conditions in the CONNECT BY clause, but the c column of Y has already appeared as c in the path leading to X.
When trying to understand CONNECT BY queries, I find it helpful to display the results of SYS_CONNECT_BY_PATH. For example, the following query
SELECT  SUBSTR ( SYS_CONNECT_BY_PATH ( '('  || p || ',' || c || ')'
                             , ' -> '
            , 5
            )          AS path
,     CONNECT_BY_ISCYCLE       AS iscycle
FROM     t
CONNECT BY NOCYCLE     PRIOR c     = p
;given your sample data, produces this output:
PATH                                   ISCYCLE
(4,5)                                        0
(4,5) -> (5,4)                               1
(4,5) -> (5,4) -> (4,6)                      1
(4,6)                                        0
(4,6) -> (6,4)                               1
(4,6) -> (6,4) -> (4,5)                      1
(5,4)                                        0
(5,4) -> (4,5)                               1
(5,4) -> (4,6)                               1
(6,4)                                        0
(6,4) -> (4,5)                               1
(6,4) -> (4,6)                               1If you find it helpful to add other columns (p, c, LEVEL, or the CONNECT_BY_ROOTs) then add them. I feel it's easy enough to find all those things from the display above.
I wonder if you're thinking the rows in t as vertices. Don't: you'll understand CONNECT BY queries much better if you think of your rows as edges. Think of the individual c values as vertices.
943276 wrote:
Thanks Solomon. From what you say I understand that connect_by_iscycle value = 1 is inherited at all lower levels.No. Remember what the manual says: CONNECT_BY_ISCYCLE means that a node has a child which has already appeared in the path as its ancestor. That node may have other children that either do not hove children (and therefore can't have children that are their own ancestors), or have children whose children are not their own ancestors.
If you add these rows to your sample data:
INSERT INTO t (p, c) VALUES ( 5, 11);
INSERT INTO t (p, c) VALUES ( 5, 12);
INSERT INTO t (p, c) VALUES (12, 13);then the output will have 28 rows, including these 6:
PATH                                             ISCYCLE
(4,6)                                                  0
(4,6) -> (6,4)                                         1
(4,6) -> (6,4) -> (4,5)                                1
(4,6) -> (6,4) -> (4,5) -> (5,12)                      0
(4,6) -> (6,4) -> (4,5) -> (5,12) -> (12,13)           0
(4,6) -> (6,4) -> (4,5) -> (5,11)                      0Notice that the last 3 rows of this output have iscycle=0, even though they are descended from two rows that have iscycle=1.
But I really don't get the results. Let me rewrite points (a)-(h). as I understand it, they form the following possible paths:
4 -> 5 -> 4 [loop detected] -> 6  in result set
4 -> 5 -> 4 [loop detected] -> 5 
4 -> 6 -> 4 [loop detected] -> 5  in result set
4 -> 6 -> 4 [loop detected] -> 6
5 -> 4 -> 5 [loop detected] -> 4
5 -> 4 -> 6 -> 4 [loop detected]  not in result set
6 -> 4 -> 5 -> 4 [loop detected]  not in result set
6 -> 4 -> 6 [loop detected] -> 4 Why those paths are allowed
4          6 4-->5-->4-->6                 1          3
4          5 4-->6-->4-->5                 1          3even in those paths the loop is detected earlier than in paths not present in result set?What query is producing those results?