Oracle analytic function window defined by data

Question

I have a table which represents a line-by-line dump of the data read from a particular text file format. Each line may represent a "master" or a "detail" line, indicated via rec_type code. I'd like to write a query that gets the "master" lines alongside the associated detail lines. I've come up with something that does the job, but it seems a bit hackish and am interested in better ways if any.

CREATE TABLE mdtest
 (rec_seq  NUMBER        PRIMARY KEY
 ,rec_type VARCHAR2(3)   NOT NULL
 ,rec_data VARCHAR2(100) NOT NULL);

INSERT INTO mdtest VALUES (1, '100', 'Bill Jones');
INSERT INTO mdtest VALUES (2, '200', '20080115,100.25');
INSERT INTO mdtest VALUES (3, '100', 'John Smith');
INSERT INTO mdtest VALUES (4, '200', '20090701,80.95');
INSERT INTO mdtest VALUES (5, '200', '20091231,110.35');

Desired Result:

SEQ_EMP  EMP_NAME    SEQ_DATA  EMP_DATA
=======  ==========  ========  ===============
      1  Bill Jones         2  20080115,100.25
      3  John Smith         4  20090701,80.95
      3  John Smith         5  20091231,110.35

Assumptions:

• records are processed in sequence of rec_seq
• first record type is a "100"
• each "100" record has 1 or more "200" records following

Note: this is for Oracle 9i, however we should be upgrading to 11g R1 this year.

Answer 1

Here's what I have so far:

SELECT seq_emp 
      ,SUBSTR(emp_seq_name,10) emp_name 
      ,seq_data 
      ,emp_data 
FROM  (SELECT MAX(CASE WHEN rec_type = '100' THEN rec_seq END) 
              OVER (ORDER BY rec_seq 
                    ROWS BETWEEN UNBOUNDED PRECEDING 
                             AND CURRENT ROW) seq_emp 
             ,MAX(CASE 
                  WHEN rec_type = '100' 
                  THEN TO_CHAR(rec_seq,'fm00000000') || '|' || rec_data 
                  END) 
              OVER (ORDER BY rec_seq 
                    ROWS BETWEEN UNBOUNDED PRECEDING 
                             AND CURRENT ROW) emp_seq_name 
             ,rec_seq seq_data 
             ,rec_type 
             ,rec_data emp_data 
       FROM   mdtest) 
WHERE  rec_type = '200' 
ORDER BY seq_data; 

As you can see, I'm using the MAX reporting analytic function with a window starting from the top of the set down to the current row, to get the relevant "100" record for the current "200" record; then in the outer query I'm discarding the unneeded "100" records.

To get emp_name, I've had to append the rec_seq with the data so that the MAX function still picks the correct header record; then in the outer query I chop the rec_seq off.

I've played with other analytic functions and syntax including FIRST_VALUE and the KEEP syntax, but neither of these seem to make this job simpler; the difficulty is that the window is defined by the value of rec_type instead of being some constant offset.

Answer 2

In a desire for simplicity, do you think it's worth loading each record type into a separate import table before processing?

create table mdtest100 as select * from mdtest where rec_type = 100;

create table mdtest200 as select * from mdtest where rec_type = 200;

with mdtest_detail as
    (
    select
        (select max(m.rec_seq) from mdtest100 m 
         where m.rec_seq < r200.rec_seq) master_rec_seq,
        r200.* 
    from 
        mdtest200 r200
    )
select
    m.rec_seq seq_emp,
    m.rec_data emp_name,
    d.rec_seq seq_data,
    d.rec_data emp_data
from
    mdtest_detail d
        inner join mdtest100 m on m.rec_seq = d.master_rec_seq
order by
    seq_emp,
    seq_data;


    SEQ_EMP  EMP_NAME    SEQ_DATA   EMP_DATA          
    1        Bill Jones  2          20080115,100.25          
    3        John Smith  4          20090701,80.95          
    3        John Smith  5          20091231,110.35          

This might lend itself to a more maintainable solution and will let you parse out and validate the comm-separated EMP_DATA field separately.

Just a thought - sorry if you're only looking for an Analytics solution.