[8i] Need help updating/fixing a workday table

Question

I'm working in an old database:
Oracle8i Enterprise Edition Release 8.1.7.2.0 - Production
PL/SQL Release 8.1.7.2.0 - Production
CORE 8.1.7.0.0 Production
TNS for HPUX: Version 8.1.7.2.0 - Production
NLSRTL Version 3.4.1.0.0 - Production
I'm trying to fix a workday calendar, and I'm not exactly sure how to do it. I'm used to querying data, not making changes to the data...
Here's some sample data:
CREATE TABLE     test_cal
(     clndr_dt     DATE     NOT NULL
,     work_day     NUMBER(3)
,     clndr_days     NUMBER(6)
,     work_days     NUMBER(5)
INSERT INTO     test_cal
VALUES     (TO_DATE('12/30/2010','mm/dd/yyyy'), 254, 11322, 7885);
INSERT INTO     test_cal
VALUES     (TO_DATE('12/31/2010','mm/dd/yyyy'), 255, 11323, 7886);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/01/2011','mm/dd/yyyy'), 0, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/02/2011','mm/dd/yyyy'), 0, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/03/2011','mm/dd/yyyy'), 1, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/04/2011','mm/dd/yyyy'), 2, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/05/2011','mm/dd/yyyy'), 3, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/06/2011','mm/dd/yyyy'), 4, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/07/2011','mm/dd/yyyy'), 5, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/08/2011','mm/dd/yyyy'), 0, NULL, NULL);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/09/2011','mm/dd/yyyy'), 0, 11332, 7895);
INSERT INTO     test_cal
VALUES     (TO_DATE('01/10/2011','mm/dd/yyyy'), 6, 11333, 7896);Note: 12/31/2010 is the last time I am 100% sure the data in the table is good. After that, clndr_days and work_days are either missing or, at least in the case of work_days, are not correct.
This is a two-part question, since I need to fix both clndr_days and work_days. (You can probably fix them both with one statement, but I broke it into two to start, to simplify it for myself).
I only show 12 days in my sample data, but really, the table stretches back a number of years, and forward another 5 or so years.
PART 1: fixing clndr_days
I put together this query:
SELECT     clndr_dt
,     work_day
,     work_days
,     clndr_days
,     min_clndr_day + r_num     AS clndr_days_test
FROM     (
     SELECT     t.*
     ,     MIN(clndr_days)     OVER(ORDER BY clndr_dt)     AS min_clndr_day
     ,     ROW_NUMBER() OVER(ORDER BY clndr_dt)-1     AS r_num
     FROM     test_cal t
     WHERE     clndr_dt     >= TO_DATE('12/31/2010','mm/dd/yyyy')
;Which gives the right clndr_days (as clndr_days_test), but I'm not sure how to get that into the table.
This doesn't work:
UPDATE     test_cal
SET     clndr_days     =     (
                    SELECT     min_clndr_day + r_num
                    FROM     (
                         SELECT     t.*
                         ,     MIN(clndr_days)     OVER(ORDER BY clndr_dt)     AS min_clndr_day
                         ,     ROW_NUMBER() OVER(ORDER BY clndr_dt)-1     AS r_num
                         FROM     test_cal t
                         WHERE     clndr_dt     >= TO_DATE('12/31/2010','mm/dd/yyyy')
                         ) c
                    WHERE     c.clndr_dt     = ???  -- how do I make this equal whatever the clndr_dt is for the record I'm updating?
WHERE     clndr_dt     >      TO_DATE('12/31/2010','mm/dd/yyyy')and I'm not sure how to make it work...
PART 2: Fixing work_days
Then, I can't seem to put together a query for work_days.
This is what I have so far, but it doesn't work quite right, and then it also needs to be an UPDATE statement as well:
SELECT     clndr_dt
,     work_day
,     work_days
,     clndr_days
,     min_work_day + r_num     AS work_days_test --this isn't right, when work_day is 0 work_days_test should be the previous work_day not the minimum work_day
FROM     (
     SELECT     t.*
     ,     MIN(work_days)     OVER ( ORDER BY clndr_dt)     AS min_work_day
     ,     CASE
               WHEN     work_day     <> 0
               THEN     ROW_NUMBER()     OVER ( PARTITION BY     CASE
                                                  WHEN     work_day <> 0
                                                  THEN     1
                                                  ELSE     2
                                             END
                                     ORDER BY     clndr_dt
               ELSE     0
          END               AS r_num
     FROM     test_cal t
     WHERE     clndr_dt     >= TO_DATE('12/31/2010','mm/dd/yyyy')
;(When I tried using LAG, that didn't work either, because you can have more than 1 non-work day in a row, and so it only works for the first non-work day to have the correct work_days, and then it's back to being wrong again)
This is what I want my table to look like in the end:
CLNDR_DT                   WORK_DAY       WORK_DAYS CLNDR_DAYS_TEST
12-31-2010 00:00:00         255.000       7,886.000    11,323.000
01-01-2011 00:00:00            .000       7,886.000    11,324.000
01-02-2011 00:00:00            .000       7,886.000    11,325.000
01-03-2011 00:00:00           1.000       7,887.000    11,326.000
01-04-2011 00:00:00           2.000       7,888.000    11,327.000
01-05-2011 00:00:00           3.000       7,889.000    11,328.000
01-06-2011 00:00:00           4.000       7,890.000    11,329.000
01-07-2011 00:00:00           5.000       7,891.000    11,330.000
01-08-2011 00:00:00            .000       7,891.000    11,331.000
01-09-2011 00:00:00            .000       7,891.000    11,332.000
01-10-2011 00:00:00           6.000       7,892.000    11,333.000(sorry about all the extra decimal places)
Edited by: user11033437 on Jan 11, 2012 1:40 PM

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statements, the very clear desired output, and your attempts; that's all very helpful. It would also be helpful to explain how you get the results you want from the sample data. There's always a chance someone might coincidentally get the right results for the wrong reasons with the small set of sample data, but would cause you to get wrong results with your full data.
It looks like clndr_days is the total number of days since some conventional starting point (maybe when the company started), and that work_days is the total number of work days from some starting point. It looks like work_day is 0 when the day is not a working day, and otherwise, it's the total number of working days so far in the calendar year. Is that right?
You said the data after December 31, 2010 isn't reliable. Did you mean that two columns (clndr_days and work_days) are unreliable, but the rest of the data is reliable? That is, are you sure there is exactly one row per day even after 2010, and that the work_day column is accurate? If so, you can do this:
UPDATE       test_cal     m
SET       (clndr_days, work_days) =
     SELECT     MIN (clndr_days) + COUNT (*) - 1          -- clndr_days
     ,     MIN (work_days)  + COUNT ( CASE
                                           WHEN  work_day > 0
                                AND   clndr_dt > TO_DATE ('12/31/2010', 'MM/DD/YYYY')
                                THEN  1
                            END
                             )               -- work_days
     FROM     test_cal
     WHERE     clndr_dt     BETWEEN     TO_DATE ('12/31/2010', 'MM/DD/YYYY')
                    AND     m.clndr_dt
WHERE     clndr_dt     > TO_DATE ('12/31/2010', 'MM/DD/YYYY')
;You were right: it is possible to set both columns at the same time.
It's a shame that you're using Oracle 8.1. The MERGE command, which was new in Oracle 9.1, is a lot clearer and more efficient. If you could use MERGE, you could basically use the code you posted as the core of a MERGE statement.
The UPDATE statement above assumes that, for the inaccurate days, clndr_days and work_days will never be lower than the last accurate value. If that's not the case, the solution is a little more complicated. You could avoid that problem, and make the statment faster as well, by simply hard-coding the last accurate values of clndr_days and work_days in the UPDATE statment, where I used MIN. (This sounds like one situation where efficientcy isn't a big deal, however. You'll probably never do this exact UPDATE statement again, so whether it runs in 1 second or 10 minutes might not matter much.)
Sorry, I don't have an Oracle 8 database to test this. It works in Oracle 9.2, and I don't believe it uses any features that aren't available in 8.1.