A function in a subquery is call too many times.

Question

Dear all,
I'm struggling to understand why a function in a subquery is called too many times.
Let me explain with an example:
create or replace function all_emp (v_deptno in number)
return varchar2
as
v_all_emp varchar2(2000);
begin
    dbms_output.put_line ('function called');
    for i in (select * from emp where deptno = v_deptno) loop
        v_all_emp := v_all_emp || i.ename || '; ';
    end loop;
return v_all_emp;
end;
-- running just the subquery, calls the function all_emp only 4 times (once for each row in table dept)
select
    d.deptno,
    d.dname,
    all_emp(d.deptno) f_all_emp
    from dept d;
-- running the whole query, using regexp to split the value of f_all_emp into separate fields, causes that function all_emp is called 28 times, thus 6 times for each row!!
select tmp.*,
regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
regexp_substr(f_all_emp,'[^;]*',1,11) emp6
from
    (select
    d.deptno,
    d.dname,
    all_emp(d.deptno) f_all_emp
    from dept d) tmp
;I don't understand why Oracle is calling my function 28 times in this example, 4 times should be sufficient.
Is there a way to force that the subquery is materialized first?
Little background:
Above function / query is of course a simple example.
Actually I have pretty complex function, embedding in a subquery.
The subquery is already slow (2 min to run), but when I want to split the result of the funciton in multiple (approx 20) fields it's over an hour due to above described behaviour.

Answer 1

Optimizer merges in-line view and query results in:
select  d.deptno,
        d.dname,
        all_emp(d.deptno) f_all_emp
        regexp_substr(all_emp(d.deptno),'[^;]*',1,1) emp1,
        regexp_substr(all_emp(d.deptno),'[^;]*',1,3) emp2,
        regexp_substr(all_emp(d.deptno),'[^;]*',1,5) emp3,
        regexp_substr(all_emp(d.deptno),'[^;]*',1,7) emp4,
        regexp_substr(all_emp(d.deptno),'[^;]*',1,9) emp5,
        regexp_substr(all_emp(d.deptno),'[^;]*',1,11) emp6
  from  dept d
/That's why function is called 28 times. We can see it from explain plan:
SQL> explain plan for
  2  select tmp.*,
  3          regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
  4          regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
  5          regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
  6          regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
  7          regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
  8          regexp_substr(f_all_emp,'[^;]*',1,11) emp6
  9    from  (
10           select  d.deptno,
11                   d.dname,
12                   all_emp(d.deptno) f_all_emp
13             from  dept d
14          ) tmp
15  /
Explained.
SQL> @?\rdbms\admin\utlxpls
PLAN_TABLE_OUTPUT
Plan hash value: 3383998547
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |      |     4 |    52 |     3   (0)| 00:00:01 |
|   1 |  TABLE ACCESS FULL| DEPT |     4 |    52 |     3   (0)| 00:00:01 |
8 rows selected.
SQL>  If we use NO_MERGE hint:
SQL> select  /*+ NO_MERGE(tmp) */
  2          tmp.*,
  3          regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
  4          regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
  5          regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
  6          regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
  7          regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
  8          regexp_substr(f_all_emp,'[^;]*',1,11) emp6
  9    from  (
10           select  d.deptno,
11                   d.dname,
12                   all_emp(d.deptno) f_all_emp
13             from  dept d
14          ) tmp
15  /
    DEPTNO DNAME          F_ALL_EMP  EMP1       EMP2       EMP3       EMP4       EMP5       EMP6
        10 ACCOUNTING     CLARK; KIN CLARK       KING       MILLER
                          G; MILLER;
        20 RESEARCH       SMITH; JON SMITH       JONES      SCOTT      ADAMS      FORD
                          ES; SCOTT;
                           ADAMS; FO
                          RD;
        30 SALES          ALLEN; WAR ALLEN       WARD       MARTIN     BLAKE      TURNER     JAMES
                          D; MARTIN;
    DEPTNO DNAME          F_ALL_EMP  EMP1       EMP2       EMP3       EMP4       EMP5       EMP6
                           BLAKE; TU
                          RNER; JAME
                          S;
        40 OPERATIONS
function called
function called
function called
function called
function called
function called
SQL> explain plan for
  2  select  /*+ NO_MERGE(tmp) */
  3          tmp.*,
  4          regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
  5          regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
  6          regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
  7          regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
  8          regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
  9          regexp_substr(f_all_emp,'[^;]*',1,11) emp6
10    from  (
11           select  d.deptno,
12                   d.dname,
13                   all_emp(d.deptno) f_all_emp
14             from  dept d
15          ) tmp
16  /
Explained.
SQL> @?\rdbms\admin\utlxpls
PLAN_TABLE_OUTPUT
Plan hash value: 2317111044
| Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT   |      |     4 |  8096 |     3   (0)| 00:00:01 |
|   1 |  VIEW              |      |     4 |  8096 |     3   (0)| 00:00:01 |
|   2 |   TABLE ACCESS FULL| DEPT |     4 |    52 |     3   (0)| 00:00:01 |
9 rows selected.
SQL> Not sure why function is executed 6 and not 4 times. What we actually want is to materialize in-line view:
SQL> with tmp as (
  2               select  /*+ materialize */
  3                       d.deptno,
  4                       d.dname,
  5                       all_emp(d.deptno) f_all_emp
  6                 from  dept d
  7              )
  8  select  tmp.*,
  9          regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
10          regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
11          regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
12          regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
13          regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
14          regexp_substr(f_all_emp,'[^;]*',1,11) emp6
15    from  tmp
16  /
    DEPTNO DNAME          F_ALL_EMP  EMP1       EMP2       EMP3       EMP4       EMP5       EMP6
        10 ACCOUNTING     CLARK; KIN CLARK       KING       MILLER
                          G; MILLER;
        20 RESEARCH       SMITH; JON SMITH       JONES      SCOTT      ADAMS      FORD
                          ES; SCOTT;
                           ADAMS; FO
                          RD;
        30 SALES          ALLEN; WAR ALLEN       WARD       MARTIN     BLAKE      TURNER     JAMES
                          D; MARTIN;
    DEPTNO DNAME          F_ALL_EMP  EMP1       EMP2       EMP3       EMP4       EMP5       EMP6
                           BLAKE; TU
                          RNER; JAME
                          S;
        40 OPERATIONS
function called
function called
function called
function called
SQL> explain plan for
  2  with tmp as (
  3               select  /*+ materialize */
  4                       d.deptno,
  5                       d.dname,
  6                       all_emp(d.deptno) f_all_emp
  7                 from  dept d
  8              )
  9  select  tmp.*,
10          regexp_substr(f_all_emp,'[^;]*',1,1) emp1,
11          regexp_substr(f_all_emp,'[^;]*',1,3) emp2,
12          regexp_substr(f_all_emp,'[^;]*',1,5) emp3,
13          regexp_substr(f_all_emp,'[^;]*',1,7) emp4,
14          regexp_substr(f_all_emp,'[^;]*',1,9) emp5,
15          regexp_substr(f_all_emp,'[^;]*',1,11) emp6
16    from  tmp
17  /
Explained.
SQL> @?\rdbms\admin\utlxpls
PLAN_TABLE_OUTPUT
Plan hash value: 634594723
| Id  | Operation                  | Name                       | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT           |                            |     4 |  8096 |     5   (0)| 00:00:01 |
|   1 |  TEMP TABLE TRANSFORMATION |                            |       |       |            |          |
|   2 |   LOAD AS SELECT           |                            |       |       |            |          |
|   3 |    TABLE ACCESS FULL       | DEPT                       |     4 |    52 |     3   (0)| 00:00:01 |
|   4 |   VIEW                     |                            |     4 |  8096 |     2   (0)| 00:00:01 |
|   5 |    TABLE ACCESS FULL       | SYS_TEMP_0FD9D6603_20255AE |     4 |    52 |     2   (0)| 00:00:01 |
PLAN_TABLE_OUTPUT
12 rows selected.
SQL> However, hint MATERIALIZE is an undocumented hint.
SY.