Is there an easier/simpler way of obtaining this result ?

Question

Good morning (afternoon to you, BluShadow),
I obtained the following correct, as desired, output (derived from the EMP table):
D10     D20     D30     PREZ    MGRS    ANALS   SALESM  CLERKS
CLARK   JONES   WARD    KING    BLAKE   FORD    ALLEN   ADAMS
KING    FORD    TURNER          CLARK   SCOTT   MARTIN  JAMES
MILLER  ADAMS   ALLEN           JONES           TURNER  MILLER
        SMITH   JAMES                           WARD    SMITH
        SCOTT   BLAKE
                MARTINusing the following query:
with
   -- pivoted departments  (haven't studied the Oracle PIVOT clause yet)
   depts as
    select max(case deptno
                 when 10 then ename
               end)                                 as d10,
           max(case deptno
                 when 20 then ename
               end)                                 as d20,
           max(case deptno
                 when 30 then ename
               end)                                 as d30,
           rnd
      from (
            select deptno,
                   ename,
                   row_number() over (partition by deptno
                                          order by deptno)  rnd
              from emp
     group by rnd
     order by rnd
   -- pivoted jobs
   jobs as
    select max(case job
                 when 'CLERK'         then ename
               end)                                 as Clerks,
           max(case job
                 when 'PRESIDENT'     then ename
               end)                                 as Prez,
           max(case job
                 when 'MANAGER'       then ename
               end)                                 as Mgrs,
           max(case job
                 when 'ANALYST'       then ename
               end)                                 as Anals,
           max(case job
                 when 'SALESMAN'      then ename
               end)                                 as SalesM,
           rnj
      from (
            select job,
                   ename,
                   row_number() over (partition by job
                                          order by ename)   as rnj
              from emp
     group by rnj
     order by rnj
select d10,
       d20,
       d30,
       Prez,
       Mgrs,
       Anals,
       SalesM,
       Clerks
  from depts a full outer join jobs b
                            on a.rnd = b.rnj
order by rnj, rnd;which takes a total of 5 selects to get there.
I was trying to find a query that would be, hopefully simpler, easier and didn't require as many selects. My last attempt is the following (which is close to the desired result but doesn't get a cigar yet):
select case deptno
         when 10 then ename
       end                                 as d10,
       case deptno
         when 20 then ename
       end                                 as d20,
       case deptno
         when 30 then ename
       end                                 as d30,
       case job
         when 'CLERK'         then ename
       end                                 as Clerks,
       case job
         when 'PRESIDENT'     then ename
       end                                 as Prez,
       case job
         when 'MANAGER'       then ename
       end                                 as Mgrs,
       case job
         when 'ANALYST'       then ename
       end                                 as Anals,
       case job
         when 'SALESMAN'      then ename
       end                                 as SalesM,
       row_number() over (partition by deptno
                              order by deptno)  as rnd,
       row_number() over (partition by job
                              order by ename)   as rnj
  from emp
order by rnj;The above query gets me to this result which is encouraging but... short of the mark:
D10     D20     D30     CLERKS  PREZ    MGRS    ANALS   SALESM   RND  RNJ
                ALLEN                                   ALLEN      3    1
        ADAMS           ADAMS                                      2    1
                BLAKE                   BLAKE                      6    1
KING                            KING                               2    1
        FORD                                    FORD               1    1
        SCOTT                                   SCOTT              5    2
                JAMES   JAMES                                      5    2
CLARK                                   CLARK                      3    2
                MARTIN                                  MARTIN     2    2
                TURNER                                  TURNER     1    3
MILLER                  MILLER                                     1    3
D10     D20     D30     CLERKS  PREZ    MGRS    ANALS   SALESM   RND  RNJ
        JONES                           JONES                      3    3
                WARD                                    WARD       4    4
        SMITH           SMITH                                      4    4It uses only one SELECT statement and has all the data that needs to be displayed but, I cannot find a way of eliminating the nulls without losing either some jobs or some depts.
Your help is welcome and appreciated,
John.
PS: I'll be perfectly happy learning that there is no easier/simpler way. In other words, if the answer is simply "No, there is no simpler/easier way", please do let me know that is the case. (I ask that you be fairly sure of that though, thank you)
Edited by: 440bx - 11gR2 on Jul 25, 2010 7:19 AM - Added PS.

Answer 1

Hi, John,
You had part of the solution in each of your attempts.
Do a FULL OUTER JOIN, as in your first query, on two copies of the result set of your second query.
Using Oracle 9 features only:
WITH     got_row_numbers     AS
     select     case WHEN deptno = 10 then ename end               as d10,
                 case WHEN deptno = 20 then ename end                    as d20,
                 case WHEN deptno = 30 then ename end                    as d30,
                 case WHEN job = 'CLERK'     then ename       end        as Clerks,
                 case WHEN job = 'PRESIDENT' then ename       end        as Prez,
                 case WHEN job = 'MANAGER'   then ename       end        as Mgrs,
                 case WHEN job = 'ANALYST'   then ename       end        as Anals,
                 case WHEN job = 'SALESMAN'  then ename       end        as SalesM,
                 row_number () over ( partition by      deptno
                                        order by           NULL
                       )                           as rnd,
                 row_number () over ( partition by      job
                                        order by           ename
                       )                            as rnj
  from      emp
SELECT       MIN (d.d10)          AS d10
,        MIN (d.d20)          AS d20
,       MIN (d.d30)          AS d30
,       MIN (j.clerks)     AS clerks
,       MIN (j.prez)          AS prez
,       MIN (j.mgrs)          AS mgrs
,       MIN (j.anals)          AS anals
-- ,        MIN (j.salesm)     AS salesm
FROM            got_row_numbers     d
FULL OUTER JOIN     got_row_numbers     j     ON     d.rnd     = j.rnj
GROUP BY  NVL (d.rnd, j.rnj)
ORDER BY  NVL (d.rnd, j.rnj)
;I've been trying to think of a good name for this kind of query where the items one the n-th row have nothing in common except that they are on the n-th row. For lack of anything better, I call it a Prix Fixe Query , because it resembles the menus where, for a fixed price, yuu can choose different options for each course:
Appetizer     Soup          Main Course     Desert
Pakora          Coconut          Aloo Gobi     Galabjamun
Salad          Lentil          Bharta          Halwa
Samosa                    Dhosa          Kulfi
                    Saag PaneerAbove, each column is sorted alphabeticlly. There is nothing but pure coincidence linking 'Pakora' with 'Coconut' or "Aloo Ghobi' with 'Galabjamun': they just happen the be the first items, in alphabetic order, in their respective columns.
You may notice that I used
"PARTITION BY deptno ORDER BY NULL " where you used
"PARTITION BY deptno ORDER BY deptno "
It never makes sense to ORDER BY anything in the PARTITION BY list. Each distinct item in the PARTITION BY list is a world of its own. You will only sort things that are identical with respect to the PARTITION BY list. That is, if the system has to decide whether to give 'CLARK' or 'KING' the lower ROW_NUMBER in deptno=10, deptno itself will not help the decision: the deptno for both will necessarily be the same. There's no syntax error, it just doesn't do any good, and the order of the output will be arbitrary.
There would be a syntax error if you omitted the ORDER BY clause: ROW_NUMBER must be done in the context of some ordering. If you really want the order to be arbitrary, then be clear about it. ORDER BY a constant (such as NULL) so that nobody reading the query quickly is fooled into thinking you really are ordering by something. (A comment would be helpful in that case, also.)
You probably want to "ORDER BY ename", or something meaningful, in this case.
Edited by: Frank Kulash on Jul 25, 2010 2:41 PM
Added digression of "PARTITION BY x ORDER BY x"