Query Question - Linking in Time Span Data to Point-in-Time References

Question

I am trying to pull historical data from a time-span table of records, using another point-in-time reference. I have been unsuccessful.
I've thrown some sample tables and data together to illustrate what I am seeking:
Create Table EmployeeInfo (
       id Number(10,0),               -- Employee ID Number
       Name VarChar2(32 Byte),        -- Employee Name
       CurrentShift VarChar2(2 Byte)  -- Current Employee Shift
Create Table ShiftChanges (
       id Number(10,0),               -- Employee ID that this shift change record is for
       ChangeDate Date,               -- Date that this Change Took Place
       OldShift VarChar2(2 Byte),
       NewShift VarChar2(2 Byte)
Create Table TimeCard(
       id Number(10,0),               -- Employee ID Number for this Timecard
       WorkDay Date,                  -- What Day is this Timecard for?
       Hours Float(63)                -- Number of Hours worked.
COMMIT;
INSERT INTO EmployeeInfo VALUES (100,'John Doe','Days')
INSERT INTO EmployeeInfo VALUES (101,'Jane Doe','Days');
INSERT INTO TimeCard VALUES (100, to_date('01012010','ddmmyyyy'), 10.5);
INSERT INTO TimeCard VALUES (101, to_date('01012010','ddmmyyyy'), 10);
INSERT INTO TimeCard VALUES (100, to_date('02012010','ddmmyyyy'), 9);
INSERT INTO TimeCard VALUES (101, to_date('02012010','ddmmyyyy'), 10.5);
INSERT INTO TimeCard VALUES (100, to_date('03012010','ddmmyyyy'), 8);
INSERT INTO TimeCard VALUES (101, to_date('03012010','ddmmyyyy'), 7);
INSERT INTO ShiftChanges VALUES (100, to_date('02012010','ddmmyyyy'), 'Nights', 'Days');
COMMIT;I could do a query such as:
SELECT TC.id,
       EM.Name,
       TC.Workday,
       EM.CurrentShift AS Shift
FROM   TimeCard TC,
       EmployeeInfo EM
WHERE  (TC.ID=EM.ID(+));But it will not give me the historical shift, only the current. Since John Doe changed shifts on Jan 2 from Nights to Days, the above query would not reflect it.
I have attempted to join the historical table using a less-than operator. This doesn't work since more than one row can be returned.
SELECT TC.id,
       EM.Name,
       TC.Workday,
       SC.NewShift AS Shift
FROM   TimeCard TC,
       EmployeeInfo EM,
       ShiftChanges SC
WHERE  (TC.ID=EM.ID(+)) AND
       (TC.ID=SC.ID(+) AND TC.WORKDAY<=SC.WORKDAY(+));The problem stems from the fact that you need to examine multiple records in one table in order to obtain the correct historical data for the other.
The following SQL script {color:green}does work{color} - if the values are defined ahead of time.
DEFINE EMPID=101;
DEFINE CHGDATE=to_date('01/01/2010','dd/mm/yyyy');
SELECT SQ.Shift
FROM   (SELECT  ShiftChanges.NewShift AS Shift,
                RANK() OVER (ORDER BY ShiftChanges.ChangeDate DESC) R
        FROM    ShiftChanges
        WHERE   ShiftChanges.id = &EMPID AND
                ShiftChanges.ChangeDate <= &CHGDATE
        ) SQ
WHERE R = 1However, I have been unsuccessful in adapting it to the example tables I provided. If I insert the query as an inline subquery* in the select statement, it won't work, since the criteria is nested two levels down, and I can only get parent values within the first level.
I haven't thought of a way I can do this using a nested subquery - I keep running into that problem - how do you link in Time-Span data with a point-in-time reference.
Any ideas / enlightening thoughts?
Thank You
{size:8}_SELECT * FROM V$VERSION information:_
Oracle9i Enterprise Edition Release 9.2.0.8.0 - Production
PL/SQL Release 9.2.0.8.0 - Production
"CORE     9.2.0.8.0     Production"
TNS for 32-bit Windows: Version 9.2.0.8.0 - Production
NLSRTL Version 9.2.0.8.0 - Production
{size}
user10879184 re-wrote the original post on 29-Mar-2010 1:21 PM to reflect suggestions made by poster.

Answer 1

I changed the tables to hold VARCHAR2(6) shift information and your query to reference the ChangeDate column.
I also assume you meant to link a TimeCard entry with the most previous ShiftChange entry rather than one that occurs after the time worked (TC.WORKDAY>=SC.CHANGEDATE(+) rather than TC.WORKDAY<=SC.CHANGEDATE(+)):
Another issue you don't take into account the case where there has been no shift change. In that event you need the current shift from the EmployeeInfo table which is taken care of with the NVL function.
You correctly note that multiple shift change records will result in duplicate TimeCard records though your test data fails to check for that condition. I added this row:
SQL> INSERT INTO ShiftChanges VALUES (100, to_date('03012010','ddmmyyyy'), 'Days', 'Nights'); Now it is apparent
SQL> SELECT TC.id,
  2         EM.Name,
  3         TC.Workday,
  4         NVL(SC.NewShift,EM.CurrentShift) AS Shift
  5  FROM   TimeCard TC,
  6         EmployeeInfo EM,
  7         ShiftChanges SC
  8  WHERE  TC.ID=EM.ID(+) AND
  9         TC.ID=SC.ID(+) AND TC.WORKDAY>=SC.CHANGEDATE(+);
        ID NAME                             WORKDAY   SHIFT
       100 John Doe                         01-JAN-10 Days
       100 John Doe                         02-JAN-10 Days
       100 John Doe                         03-JAN-10 Days
       100 John Doe                         03-JAN-10 Nights
       101 Jane Doe                         01-JAN-10 Days
       101 Jane Doe                         02-JAN-10 Days
       101 Jane Doe                         03-JAN-10 Days
7 rows selected.The reason for the duplicate Jan3 time is the two shift change records. We are matching both ShiftChange records that occured before the 3rd.
We also see that your test data has John starting out on Days and then shifting from Nights to Days on the 2nd for no real change. Fixing that:
SQL> DELETE from ShiftChanges where ID = 100;
2 rows deleted.
SQL> INSERT INTO ShiftChanges VALUES (100, to_date('02012010','ddmmyyyy'), 'Days', 'Nights');
1 row created.
SQL> INSERT INTO ShiftChanges VALUES (100, to_date('03012010','ddmmyyyy'), 'Nights', 'Days');
1 row created.Then:
SQL> SELECT TC.id,
  2         EM.Name,
  3         TC.Workday,
  4         NVL(SC.NewShift,EM.CurrentShift) AS Shift
  5  FROM   TimeCard TC,
  6         EmployeeInfo EM,
  7         (SELECT ID, Workday, ChangeDate, NewShift
  8          FROM
  9                 (SELECT TC2.ID, TC2.Workday, SC2.ChangeDate, SC2.Newshift,
10                         MAX (SC2.ChangeDate) KEEP (DENSE_RANK LAST ORDER BY SC2.ChangeDate)
11                                              OVER (PARTITION BY SC2.ID, TC2.Workday) AS Max_ChangeDate
12                  FROM   ShiftChanges SC2,
13                         TimeCard TC2
14                  WHERE  TC2.Workday >= SC2.ChangeDate
15                  AND    TC2.ID = SC2.ID
16                  )
17          WHERE   ChangeDate = Max_ChangeDate
18          ) SC
19  WHERE  TC.ID=EM.ID(+) AND
20         TC.ID=SC.ID(+) AND
21         TC.Workday=SC.ChangeDate(+) AND
22         TC.Workday = SC.Workday(+);
        ID NAME                             WORKDAY   SHIFT
       100 John Doe                         01-JAN-10 Days
       100 John Doe                         02-JAN-10 Nights
       100 John Doe                         03-JAN-10 Days
       101 Jane Doe                         01-JAN-10 Days
       101 Jane Doe                         02-JAN-10 Days
       101 Jane Doe                         03-JAN-10 Days
6 rows selected.The inline view finds the greatest ChangeDate less than the TimeCard Workday for each Workday.