Merging two rows based on a sequence_number

Question

Hi all
Suppose that I have a table with the following structure:
UserID          NUMBER          PK
SequenceID     NUMBER          NOT NULL
Value1          NUMBER          NULL
Value2          VARCHAR2          NULL
Value3          DATE          NULL
And, this table, is populated as examplified below:
UserID     SequenceID     Value1     Value2     Value3
1     1          10     NULL     'TEST'     
1     2          15     NULL     01/01/2011
1     3          NULL     'TEST2'     02/01/2011
2     4          20     NULL     NULL
2     5          12     'ABC'     NULL
2     6          NULL     'AB'     NULL
What I want to do is merge these rows, "grouping" then by UserID (That is the PK), on a way that the last major sequence ID always takes precedence from the previous one... So, the query should returns something like:
UserID     SequenceID     Value1     Value2     Value3
1     3          15     'TEST2'     02/01/2011
2     6          12     'AB'     NULL
Explanations:
As you can see, for each UserID, I´d like to get a row with the last filled value for each columns...
I can´t use the MAX function because, to UserID 2, I will get unexpected values on column Value2 (I would get 'ABC' insted of 'AB') and column Value1 (I would get 20 instead of 12).
How is the best way (best performance) to archieve this?
Thanks a lot!!!

Answer 1

Hi,
Maybe this will help:
select userid,
       max(sequenceid) keep (dense_rank last order by sequenceid) sequenceid,
       max(value1) keep (dense_rank last order by case when value1 is not null then sequenceid end nulls first) value1,
       max(value2) keep (dense_rank last order by case when value2 is not null then sequenceid end nulls first) value2,
       max(value3) keep (dense_rank last order by case when value3 is not null then sequenceid end nulls first) value3
  from t
group by userid;Sample execution:
SQL> with t as (
  2  select 1 userid, 1 sequenceid, 10 value1, NULL value2, 'TEST' value3 from dual union all
  3  select 1, 2, 15, NULL, '01/01/2011' from dual union all
  4  select 1, 3, NULL, 'TEST2', '02/01/2011' from dual union all
  5  select 2, 4, 20, NULL, NULL from dual union all
  6  select 2, 5, 12, 'ABC', NULL from dual union all
  7  select 2, 6, NULL, 'AB', NULL from dual
  8  ) -- end of test data
  9  select userid,
10         max(sequenceid) keep (dense_rank last order by sequenceid) sequenceid,
11         max(value1) keep (dense_rank last order by case when value1 is not null then sequenceid end nulls first) value1,
12         max(value2) keep (dense_rank last order by case when value2 is not null then sequenceid end nulls first) value2,
13         max(value3) keep (dense_rank last order by case when value3 is not null then sequenceid end nulls first) value3
14    from t
15   group by userid;
    USERID SEQUENCEID     VALUE1 VALUE2 VALUE3
         1          3         15 TEST2  02/01/2011
         2          6         12 AB     You can take the result of this SQL and populate the new version of table with an Insert .. Select statement.
Edited by: fsitja on Feb 25, 2011 2:57 PM Fixed it and removed test code that went "live" by mistake