Date/Time sequence generation

Question

Hello,
NB: database is currently Oracle 8i (circumstances), migration to Oracle 9i is planned but not imminent.
I have timestamped data logged to a database. The logging of the data
occurs every second, but to reduce space constraints unless data falls
outside a deadband it is not logged to the database.
I'm looking for routines or methodology to fill in the missing data using
SQL or PL/SQL (I have already written Java to do this on the client
side but my gut feeling is that server side manipulation of the data in
PL/SQL or even straight SQL would be more efficient)
For example, given the following data where the time stamp is
YYYYMMDDHH24MISS:
Timestamp TagValue
20030328080000 -10
20030328080005 0
20030328080010 -10
20030328080020 -20
I would need to produce:
Timestamp TagValue
20030328080000 -10
20030328080001 -10
20030328080002 -10
20030328080003 -10
20030328080004 -10
20030328080005 0
20030328080006 0
20030328080007 0
20030328080008 0
20030328080009 0
20030328080010 -10
20030328080011 -10
20030328080012 -10
20030328080013 -10
20030328080014 -10
20030328080015 -10
20030328080016 -10
20030328080017 -10
20030328080018 -10
20030328080019 -10
20030328080020 -10
Data in this example has changed frequently but in real life the data
does not change for days. The purpose of filling in the holes in the data
is to allow complete reuse of the client side code which depends on the
data being in the 1 second format. Previously data was logged to a
custom binary file and when retrieved it was automatically put into the
1 second format. The logging is being ported to and Oracle backend
and I'm involved with updating of the client.
Long winded description boils down to:
Given two arbitrary timestamps, how do I generate all timestamps
between them using a given interval (i.e. above we use 1 second,
however request might be for 5 second, 1 minute, or 1 hour data
interval)??
Doing this on the server side will allow easy migration to the current
client and then easy update of the client as many of the routines which
currently iteratively process the data could be replaced by simple
queries (averages, sums, dates when certain events occurred, etc).
Thanks.
-- John

Answer 1

This is kind of ugly, but it does the job.
SQL> SELECT TO_CHAR(ev_dt,'yyyymmddhh24miss'),val
  2  FROM t;
TO_CHAR(EV_DT,        VAL
20030328080000        -10
20030328080005          0
20030328080010        -10
20030328080020        -20
SQL> SELECT dt,NVL(val,(SELECT val
  2                     FROM t
  3                     WHERE ev_dt = (SELECT MAX(ev_dt)
  4                                    FROM t
  5                                    WHERE ev_dt <= TO_DATE(dt,'yyyymmddhh24mi
ss')))) val
  6  FROM (SELECT TO_CHAR(start_dt + ((rownum -1)/24/60/60),'yyyymmddhh24miss')
dt
  7        FROM (SELECT MIN(ev_dt) start_dt, MAX(ev_dt) last_dt
  8              FROM t),
  9             (SELECT 1
10              FROM dba_tables
11              WHERE rownum <= (SELECT ((MAX(ev_dt) - MIN(ev_dt))*24*60*60)+2
12                               FROM t))
13        WHERE start_dt+(ROWNUM/24/60/60) <= last_dt+(1/24/60/60)) a, t b
14  WHERE TO_DATE(a.dt,'yyyymmddhh24miss') = b.ev_dt(+);
DT                    VAL
20030328080000        -10
20030328080001        -10
20030328080002        -10
20030328080003        -10
20030328080004        -10
20030328080005          0
20030328080006          0
20030328080007          0
20030328080008          0
20030328080009          0
20030328080010        -10
20030328080011        -10
20030328080012        -10
20030328080013        -10
20030328080014        -10
20030328080015        -10
20030328080016        -10
20030328080017        -10
20030328080018        -10
20030328080019        -10
20030328080020        -20
21 rows selected.No guarantees on performance on a big table.
TTFN
John