Question

Is it possible, in Oracle, to group data on the output of a user defined function? I get errors when I try to, and it best illustrated by the below example:

I am trying to interrogate results in table structure similar to below:

id   | data
1000 | {abc=123, def=234, ghi=111, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=123, def=234, ghi=222, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=123, def=434, ghi=333, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=123, def=434, ghi=444, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=123, def=634, ghi=555, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=923, def=634, ghi=666, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=923, def=434, ghi=777, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=923, def=434, ghi=888, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=923, def=234, ghi=999, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}
1000 | {abc=923, def=234, ghi=000, jkl=456, mno=567, pqr=678, stu=789, vwx=890, yza=901}

There are other columns, just not shown. The id column can have different values, but in this example, does not. In the data column, only the fields abc, def, and ghi differ, all the others are the same. Again this is only illustrative for this data example.

I have written a function to extract the value assigned to fields in the data column, and it is used in the following query:

select id
      ,extract_data(data,abc) as abc
      ,extract_data(data,def) as def
from   table

giving results:

id   | abc | def
1000 | 123 | 234
1000 | 123 | 234
1000 | 123 | 434
1000 | 123 | 434
1000 | 123 | 634
1000 | 923 | 634
1000 | 923 | 434
1000 | 923 | 434
1000 | 923 | 234
1000 | 923 | 234

For reporting purposes, I would like to be able to display the amount of each type of record. There are 6 types in the above example, and ideally the output would be:

id   | abc | def | count
1000 | 123 | 234 | 2
1000 | 123 | 434 | 2
1000 | 123 | 634 | 1
1000 | 923 | 634 | 1
1000 | 923 | 434 | 2
1000 | 923 | 234 | 2

I expected to achieve this by writing SQL like so (and I’m convinced I have done so in the past):

select id
      ,extract_data(data,abc) as abc
      ,extract_data(data,def) as def
      ,count(1)
from   table
group by id
        ,abc
        ,def

This however, will not work. Oracle is giving me an error of:

ORA-00904: “ABC”: invalid identifier
00904. 00000 – “%s: invalid identifier”

From my initial research on “the google”, I have seen that I should perhaps be grouping on the column I am passing into my user defined function. This would be due to SQL requiring all columns not part of an aggregate function needing to be part of the group by clause.

This will work for some records, however in my data example, the field ghi in the data column is different for every record , thus making the data column unique, and ruining the group by clause, as a count of 1 is given for each record.

I’ve used sybase and db2 in the past, and (setting myself up for a fall here…) I’m pretty sure in both that I was able to group by on the output of a user defined function.

I thought that there might be an issue with the naming of the columns and how they can be referenced by the group by? Referencing by column number hasn’t worked.

I’ve tried various combinations of what I have, and can’t get it to work, so I’d appreciate any insight you guys out there could give.

If you need any more information I’ll edit as required or clarify in the comments.

Thanks,
GC.

Answer 1

You should be able to group by the functions themselves, not by the aliases

select id
      ,extract_data(data,abc) as abc
      ,extract_data(data,def) as def
      ,count(*)
from   table
group by id
        ,extract_data(data,abc)
        ,extract_data(data,def) 

Note that this does not generally involve executing the function multiple times. You can see that yourself with a simple function that increments a counter in a package every time it is called

SQL> ed
Wrote file afiedt.buf

  1  create or replace package pkg_counter
  2  as
  3    g_cnt integer := 0;
  4* end;
SQL> /

Package created.

SQL> create or replace function f1( p_arg in number )
  2    return number
  3  is
  4  begin
  5    pkg_counter.g_cnt := pkg_counter.g_cnt + 1;
  6    return mod( p_arg, 2 );
  7  end;
  8  /

Function created.

There are 16 rows in the EMP table

SQL> select count(*) from emp;

  COUNT(*)
----------
        16

so when we execute a query that involves grouping by the function call, we hope to see the function executed only 16 times. And that is, in fact, what we see.

SQL> select deptno,
  2         f1( empno ),
  3         count(*)
  4    from emp
  5   group by deptno,
  6            f1( empno );

    DEPTNO  F1(EMPNO)   COUNT(*)
---------- ---------- ----------
                    1          1
        30          0          4
        20          1          1
        10          0          2
        30          1          2
        20          0          4
        10          1          1
                    0          1

8 rows selected.

SQL> begin
  2    dbms_output.put_line( pkg_counter.g_cnt );
  3  end;
  4  /
16

PL/SQL procedure successfully completed.