Misconception about the LAG function's ... functionality?

Question

So I want to look at the value of a column on a current row, and the value of the same column on the most recently entered row, prior to the current. The problem I'm encountering is that for the current row, I'm only interested in those added last month. So, if the record that I'm seeking in the LAG function is prior to last month, it gets rejected from the resultset.
create table check_lag
(filenr number, code varchar2(2), create_date date);
insert into check_lag values (1,'02',to_date('9/5/2008','MM/DD/YYYY')); -- current
insert into check_lag values (1,'01',to_date('9/1/2008','MM/DD/YYYY')); --lag record, same month
insert into check_lag values (2,'02',to_date('9/10/2008','MM/DD/YYYY'));-- current
insert into check_lag values (2,'01',to_date('8/10/2008','MM/DD/YYYY'));-- lag record prior monthSo this query's results made sense
SELECT FILENR, CODE,
       LAG( CODE ) OVER( PARTITION BY FILENR ORDER BY FILENR, CREATE_DATE ) AS PRIOR_CODE,
       CREATE_DATE
FROM   CHECK_LAG;
FILENR CODE PRIOR_CODE CREATE_DATE
1      01              9/1/2008
1      02   01         9/5/2008
2      01              8/10/2008
2      02   01         9/10/2008But as soon as I add a WHERE clause which set's a boundary around last month, I exclude a LAG record
SELECT FILENR, CODE,
       LAG( CODE ) OVER( PARTITION BY FILENR ORDER BY FILENR, CREATE_DATE ) AS PRIOR_CODE,
       CREATE_DATE
FROM   CHECK_LAG;
FILENR CODE PRIOR_CODE CREATE_DATE
1      01              9/1/2008
1      02   01         9/5/2008
2      02              9/10/2008I know that I could push this into an inline view, and provide the WHERE clause with the date range after the inline view is processed, but this is a huge table with an index on the CREATE_DATE, and so the following forces a table scan
SELECT *
FROM   ( SELECT FILENR, CODE,
                LAG( CODE ) OVER( PARTITION BY FILENR ORDER BY FILENR, CREATE_DATE ) AS PRIOR_CODE,
                CREATE_DATE
        FROM   CHECK_LAG )
WHERE  CREATE_DATE BETWEEN TO_DATE( '09/01/2008', 'MM/DD/YYYY' )
                       AND TO_DATE( '09/30/2008 23:59:59', 'MM/DD/YYYY HH24:MI:SS' )
AND    PRIOR_CODE IS NOT NULL;
FILENR CODE PRIOR_CODE CREATE_DATE
1      02   01         9/5/2008
2      02   01         9/10/2008Is that just the way things are, or am I missing out on another approach?
Thanks,
Chuck

Answer 1

Hi, Chuck,
Thanks for including the CREATE TABLE and INSERT statements.
When you use" ORDER BY <single_column>" in an analytic function, you can restrict it to a range of values relative to the ordering value on the current row.
For example, you could say
ORDER  BY create_date
RANGE  BETWEEN  60 PRECEDING
           AND  30 PRECEDINGif you only wanted to consider rows that were no more than 60 days earlier, but at least 30 days earlier.
It's a little more complicated for your problem, because you can't just hard-code numbers like 60 and 30; you have to compute them for every row.
SELECT     filenr
,     code
,     LAST_VALUE (code) OVER
     ( PARTITION BY     filenr
       ORDER BY     create_date
       RANGE BETWEEN     create_date - ADD_MONTHS ( TRUNC ( create_date
                                    , 'MM'
                               , -1
                               )     PRECEDING
          AND     create_date - TRUNC ( create_date
                             , 'MM'
                             )          PRECEDING
     ) AS prior_code
,     create_date
FROM     check_lag;You could probably get the same results using LAG, but this is exactly for what LAST_VALUE was designed.
Windowing is described in the [SQL Language Reference manual|http://download.oracle.com/docs/cd/B28359_01/server.111/b28286/functions001.htm#sthref972] in the section for "analytic functions" in general.