Looking forward to next rows

Question

hello everybody, i have a particular scenario that i want to solve with a query
here is some sample data. note dates are not sorted in table. i put it here for readabaility reason
WITH table1 AS
  SELECT 111 cid,  To_Date('1/31/2010','mm/dd/yyyy') dt, 'line' line,  345 pid   FROM dual UNION ALL
  SELECT 111 cid,  To_Date('2/1/2010','mm/dd/yyyy') dt,  'line2' line, 346 pid  FROM dual UNION all
  SELECT 111 cid,  To_Date('2/2/2010','mm/dd/yyyy') dt,  'line3' line, 347 pid  FROM dual UNION ALL
  SELECT 111 cid,  To_Date('2/3/2010','mm/dd/yyyy') dt,  'line4' line, 348 pid  FROM dual UNION ALL
  SELECT 111 cid,  To_Date('2/4/2010','mm/dd/yyyy') dt,   null   line, 349 pid FROM dual UNION ALL
  SELECT 111 cid,  To_Date('2/5/2010','mm/dd/yyyy') dt,  'line6' line, 340 pid  FROM dual UNION all
  SELECT 999 cid,  To_Date('2/28/2010','mm/dd/yyyy') dt, 'line7' line, 341 pid  FROM dual UNION ALL
  SELECT 999 cid,  To_Date('3/1/2010','mm/dd/yyyy') dt,  'line8' line, 342 pid  FROM dual UNION ALL
  SELECT 999 cid,  To_Date('3/2/2010','mm/dd/yyyy') dt,  'line9' line, 343 pid  FROM dual UNION ALL
  SELECT 999 cid,  To_Date('3/3/2010','mm/dd/yyyy') dt,  'line0' line, 344 pid FROM dual UNION ALL
  SELECT 999 cid,  To_Date('3/4/2010','mm/dd/yyyy') dt,  'line3' line, 345 pid  FROM dual UNION ALL
  SELECT 999 cid,  To_Date('3/5/2010','mm/dd/yyyy') dt,  'line2' line, 345 pid  FROM dual UNION ALL
  SELECT 444 cid,  To_Date('3/31/2010','mm/dd/yyyy') dt,  'line111' line, 346 pid  FROM dual UNION ALL
  SELECT 444 cid,  To_Date('4/1/2010','mm/dd/yyyy') dt,  'line112' line, 347 pid  FROM dual
SELECT 555 cid,  To_Date('4/2/2010','mm/dd/yyyy') dt,  'line12' line, 247 pid  FROM duali want the following output.
CID     DT            LINE     PID
111     1/31/2010     null     345   --copy line column value from 4th day
111     2/1/2010     line2     346
111     2/2/2010     line3     347
111     2/3/2010     line4     348
111     2/4/2010     null     349
111     2/5/2010     line6     340
999     2/28/2010     line3     341    --copy line column value from 4th day
999     3/1/2010     line8     342
999     3/2/2010     line9     343
999     3/3/2010     line0     344
999     3/4/2010     line3     345
999     3/5/2010     line2     345
444     3/31/2010     line111     346
444     4/1/2010     line112     347
555     4/2/2010     line12     247here is how i get to this output. i want to get all month end dates for a particular id. then i want to look forward and get the values from the 4th day following
the month end date. so for 1/31/2010, the 4th day is 2/4/2010. then i want to get the value from line column in the 4th row and copy that in the current row,
in this case 1/31/2010 line column becomes null since 2/4/2010 has null value
in the case of 444, 3/31 is month end but there is no 4th day to move forward, in this case just display the data.
in the case of 555, there is no month end date so just display the rows for 555 and no need to move forward
i try using last value analytic function but this function will skips row that are null which proceduce invalid result.
if the 4th row we trying to copy from has null values then i want those null values as display in output above.
also i am using oracle 9i 9.2 version and ignore null is not supported.
can someone help me write a query that give above output? thanks alot

Answer 1

Hi,
You're right: repeating basically the same CASE expression 8 (or is it 2 + 8 = 10) times is not a very good idea.
On each row of output, you want 8 columns from some row of the table: possibly the same row, possibly the 4th subsequent row. That sounds like a self-join. Think of joining a "base" row (containing the dt) to a "detail" row of the same table. When dt is the last day of the month, and there are at least 4 later rows, then the detail row will be the 4th later row. In all other cases, the detail row will be the same as the base row.
How can we find the detail row for a given base row? The same way we found the appropriate line value before; only instead of getting tyhe value of line, we need something that will uniquely identify the detail row. If your table has a one-column primary key, you could return that, and use it in the join condition. In the query below, I assume that the combination (cid, dt) uniquely identifies each row.
WITH     got_detail_dt     AS
     SELECT     cid, dt, line, pid
     ,     CASE
              WHEN  dt = LAST_DAY (dt)
              THEN  LEAD ( dt
                          , 4
                      , dt
                      ) OVER ( PARTITION BY  cid
                                      ORDER BY      dt
              ELSE  dt
          END          AS detail_dt
     FROM    table1
SELECT       b.cid
,       b.dt
,       d.line
,       d.pid
FROM       got_detail_dt       b
JOIN       got_detail_dt       d     ON     d.cid     = b.cid
                        AND      d.dt     = b.detail_dt
ORDER BY  cid
,            dt
;Adding more columns to the report is now just a matter of adding more columns to both SELECT clauses. (Depending on what output you will accept, you might make that just the SELECT clause of the sub-query, and use d.* in the main SELECT clause.) At any rate, the addition columns will just be simple column names, not complicated CASE expressions.