Hierarchical sql-how to get only branches with at least one not null leaves

Question

On 10gR2 we want below hierarchical query to return only the branches which at least have one not NULL leaf ;
-- drop table corporate_slaves purge ;
create table corporate_slaves (
slave_id integer primary key,
supervisor_id references corporate_slaves,
name varchar(100), some_column number );
insert into corporate_slaves values (1, NULL, 'Big Boss ', NULL);
insert into corporate_slaves values (2, 1, 'VP Marketing', NULL);
insert into corporate_slaves values (9, 2, 'Susan ', NULL);
insert into corporate_slaves values (10, 2, 'Sam ', NULL);
insert into corporate_slaves values (3, 1, 'VP Sales', NULL);
insert into corporate_slaves values (4, 3, 'Joe ', NULL);
insert into corporate_slaves values (5, 4, 'Bill ', 5);
insert into corporate_slaves values (6, 1, 'VP Engineering', NULL);
insert into corporate_slaves values (7, 6, 'Jane ', NULL);
insert into corporate_slaves values (8, 6, 'Bob' , 3);
SELECT sys_connect_by_path(NAME, ' / ') path, some_column col,  connect_by_isleaf isLeaf
  FROM corporate_slaves
CONNECT BY PRIOR slave_id = supervisor_id
START WITH slave_id IN
            (SELECT slave_id FROM corporate_slaves WHERE supervisor_id IS NULL) ;For this example wanted output is like this one since Marketing has no NOT NULL some_column leaves where as Engineering and Sales has at least one;
PATH                                                                            
/ Big Boss                                                                  
/ Big Boss  / VP Sales                                                      
/ Big Boss  / VP Sales / Joe                                                
/ Big Boss  / VP Sales / Joe  / Bill                                        
/ Big Boss  / VP Engineering                                                
/ Big Boss  / VP Engineering / Jane                                         
/ Big Boss  / VP Engineering / Bob                                           Regards.

Answer 1

Here is a slightly modified version, you can try it:
WITH SRC AS (
SELECT   SYS_CONNECT_BY_PATH(NAME ,
                             '/ ') path,
         SOME_COLUMN COL,
         CONNECT_BY_ISLEAF ISLEAF,
         CONNECT_BY_ROOT SLAVE_ID ROOT_SLAVE_ID,
         SLAVE_ID slave_id,
         CONNECT_BY_ROOT SUPERVISOR_ID SUPERVISOR_ID,
         SOME_COLUMN
    FROM CORPORATE_SLAVES
  CONNECT BY PRIOR SLAVE_ID = SUPERVISOR_ID
SELECT path, col, isleaf
FROM SRC
WHERE SUPERVISOR_ID IS NULL
AND SLAVE_ID IN (SELECT ROOT_SLAVE_ID FROM SRC WHERE SOME_COLUMN IS NOT NULL)
PATH                                COL                    ISLEAF                
/ Big Boss                                                 0                     
/ Big Boss / VP Sales                                      0                     
/ Big Boss / VP Sales/ Joe                                 0                     
/ Big Boss / VP Sales/ Joe / Bill   5                      1                     
/ Big Boss / VP Engineering                                0                     
/ Big Boss / VP Engineering/ Bob    3                      1                     
6 rows selectedI tested it for 1000 records in the source table (tested on Oracle 10 XE),
and .... the performance was a big surprise:
INSERT INTO corporate_slaves
SELECT SLAVE_ID + X SLAVE_ID,
       SUPERVISOR_ID + X SUPERVISOR_ID,
       NAME || ' ' || X NAME,
       some_column
FROM  CORPORATE_SLAVES
CROSS JOIN (
  SELECT 10*LEVEL x
  FROM DUAL
  CONNECT BY LEVEL <= 100
COMMIT;
SELECT count(*) FROM corporate_slaves;
COUNT(*)              
1010 Your query (slightly modified - removed leading space from the separator in CONNECT_BY_PATH):
set timings on;
CREATE TABLE BUBA1 AS
SELECT SYS_CONNECT_BY_PATH(NAME,
                            '/ ') path,
        some_column col,
        connect_by_isleaf isleaf
    FROM corporate_slaves
   WHERE slave_id IN (SELECT connect_by_root slave_id "slave_id"
                        FROM corporate_slaves
                       WHERE some_column IS NOT NULL
                      CONNECT BY PRIOR slave_id = supervisor_id)
  CONNECT BY PRIOR slave_id = supervisor_id
   START WITH SLAVE_ID IN
              (SELECT SLAVE_ID FROM CORPORATE_SLAVES WHERE SUPERVISOR_ID IS NULL)
CREATE TABLE succeeded.
6 095ms elapsedrewritten query:
CREATE TABLE BUBA2 AS
WITH SRC AS (
SELECT   SYS_CONNECT_BY_PATH(NAME ,
                             '/ ') path,
         SOME_COLUMN COL,
         CONNECT_BY_ISLEAF ISLEAF,
         CONNECT_BY_ROOT SLAVE_ID ROOT_SLAVE_ID,
         SLAVE_ID slave_id,
         CONNECT_BY_ROOT SUPERVISOR_ID SUPERVISOR_ID,
         SOME_COLUMN
    FROM CORPORATE_SLAVES
  CONNECT BY PRIOR SLAVE_ID = SUPERVISOR_ID
SELECT path, col, isleaf
FROM SRC
WHERE SUPERVISOR_ID IS NULL
AND SLAVE_ID IN (SELECT ROOT_SLAVE_ID FROM SRC WHERE SOME_COLUMN IS NOT NULL)
CREATE TABLE succeeded.
167ms elapsed
SELECT COUNT(*) FROM BUBA1;
COUNT(*)              
606 
SELECT COUNT(*) FROM BUBA2;
COUNT(*)              
606
SELECT COUNT(*) FROM(
  SELECT * FROM BUBA1
  INTERSECT
  SELECT * FROM BUBA2
COUNT(*)              
606  ANd now the above tests repeated for 10.000 records
truncate table  corporate_slaves;
insert into corporate_slaves values (1, NULL, 'Big Boss ', NULL);
insert into corporate_slaves values (2, 1, 'VP Marketing', NULL);
insert into corporate_slaves values (9, 2, 'Susan ', NULL);
insert into corporate_slaves values (10, 2, 'Sam ', NULL);
insert into corporate_slaves values (3, 1, 'VP Sales', NULL);
insert into corporate_slaves values (4, 3, 'Joe ', NULL);
insert into corporate_slaves values (5, 4, 'Bill ', 5);
insert into corporate_slaves values (6, 1, 'VP Engineering', NULL);
insert into corporate_slaves values (7, 6, 'Jane ', NULL);
insert into corporate_slaves values (8, 6, 'Bob' , 3);
INSERT INTO corporate_slaves
SELECT SLAVE_ID + X SLAVE_ID,
       SUPERVISOR_ID + X SUPERVISOR_ID,
       NAME || ' ' || X NAME,
       some_column
FROM  CORPORATE_SLAVES
CROSS JOIN (
  SELECT 10*LEVEL x
  FROM DUAL
  CONNECT BY LEVEL <= 1000
COMMIT;
SELECT count(*) FROM corporate_slaves;
COUNT(*)              
10010
CREATE TABLE BUBA22 AS
WITH SRC AS (
SELECT   SYS_CONNECT_BY_PATH(NAME ,
                             '/ ') path,
         SOME_COLUMN COL,
         CONNECT_BY_ISLEAF ISLEAF,
         CONNECT_BY_ROOT SLAVE_ID ROOT_SLAVE_ID,
         SLAVE_ID slave_id,
         CONNECT_BY_ROOT SUPERVISOR_ID SUPERVISOR_ID,
         SOME_COLUMN
    FROM CORPORATE_SLAVES
  CONNECT BY PRIOR SLAVE_ID = SUPERVISOR_ID
SELECT path, col, isleaf
FROM SRC
WHERE SUPERVISOR_ID IS NULL
AND SLAVE_ID IN (SELECT ROOT_SLAVE_ID FROM SRC WHERE SOME_COLUMN IS NOT NULL)
CREATE TABLE succeeded.
345ms elapsed
CREATE TABLE BUBA11 AS
SELECT SYS_CONNECT_BY_PATH(NAME,
                            '/ ') path,
        some_column col,
        connect_by_isleaf isleaf
    FROM corporate_slaves
   WHERE slave_id IN (SELECT connect_by_root slave_id "slave_id"
                        FROM corporate_slaves
                       WHERE some_column IS NOT NULL
                      CONNECT BY PRIOR slave_id = supervisor_id)
  CONNECT BY PRIOR slave_id = supervisor_id
   START WITH SLAVE_ID IN
              (SELECT SLAVE_ID FROM CORPORATE_SLAVES WHERE SUPERVISOR_ID IS NULL)
CREATE TABLE succeeded.
526 437ms elapsed
SELECT COUNT(*) FROM BUBA11;
COUNT(*)              
6006
SELECT COUNT(*) FROM BUBA22;
COUNT(*)              
6006
SELECT COUNT(*) FROM(
  SELECT * FROM BUBA11
  INTERSECT
  SELECT * FROM BUBA22
COUNT(*)              
6006 Wow.... 526 seconds vs. 0,4 seconds !!!
131500 % performance gain ;)
I have got similar results on Oracle 11.2