How to show a row has been "removed" from a table

Question

We maintain rows in a table with a version number which will determine the current set of accounts for a version. I need to be able to show the changes between the versions - when a row was ADDED, REMOVED or there was NO CHANGE. I can work out ADDED and NO_CHANGE without any problem, but I'm not sure how I can determine that a row has been REMOVED when it no longer exists in the next version.
I have provided an example piece of SQL below:
with w_acct1 as
(select 'A1' acct, 0 vers from dual
union all
select 'A2' acct, 0 vers from dual
union all
select 'A3' acct, 0 vers from dual
union all
select 'A1' acct, 1 vers from dual
union all
select 'A2' acct, 1 vers from dual
union all
select 'A1' acct, 2 vers from dual
union all
select 'A4' acct, 2 vers from dual)
select a.*,
       nvl(lead(acct) over (partition by acct order by vers desc),'NULL') ld,
       case when lead(acct) over (partition by acct order by vers desc) is null then
               'ADDED'
           when lead(acct) over (partition by acct order by vers desc) = acct then
               'NO_CHANGE'
           else
               'REMOVED'
           end add_remove
from w_acct1 a
order by vers,acctWhich gives me the following result:
ACCT VERS LD ADD_REMOVE
A1     0     NULL     NEW
A2     0     NULL     NEW
A3     0     NULL     NEW
A1     1     A1     NO_CHANGE
A2     1     A2     NO_CHANGE
A1     2     A1     NO_CHANGE
A4     2     NULL     NEW
The result I want is:
ACCT VERS LD ADD_REMOVE
A1     0     NULL     NEW
A2     0     NULL     NEW
A3     0     NULL     NEW
A1     1     A1     NO_CHANGE
A2     1     A2     NO_CHANGE
A3     1     NULL     REMOVED
A1     2     A1     NO_CHANGE
A2     2     NULL     REMOVED
A4     2     NULL     NEW
Note the REMOVED rows associated with the version even though they don't exist in the dataset for that version number.
Can this be done with analytic functions or some other cool Oracle feature I'm missing?
Regards
Richard

Answer 1

You can't know about a row being removed unless you have a record of that row somewhere, either as a copy of your old data (see example below) or you've recorded the information using delete triggers etc.
SQL> ed
Wrote file afiedt.buf
  1  with old_data as (select 1 as id, 'A' as dta from dual union all
  2                     select 2, 'B' from dual union all
  3                     select 3, 'C' from dual)
  4      ,new_data as (select 1 as id, 'A' as dta from dual union all
  5                    select 3, 'X' from dual union all
  6                    select 4, 'Y' from dual)
  7  --
  8      ,ins_upd as (select * from new_data minus select * from old_data)
  9      ,del_upd as (select * from old_data minus select * from new_data)
10      ,upd as (select id from ins_upd intersect select id from del_upd)
11      ,ins as (select id from ins_upd minus select id from upd)
12      ,del as (select id from del_upd minus select id from upd)
13  --
14  select 'Inserted' as action, null as old_id, null as old_dta, new_data.id as new_id, new_data.dta as new_dta
15  from new_data join ins on (ins.id = new_data.id)
16  union all
17  select 'Updated', old_data.id, old_data.dta, new_data.id, new_data.dta
18  from old_data join new_data on (old_data.id = new_data.id)
19                join upd on (upd.id = new_data.id)
20  union all
21  select 'Deleted', old_data.id as old_id, old_data.dta as old_dta, null as new_id, null as new_dta
22  from old_data join del on (del.id = old_data.id)
23  union all
24  select 'No Change' as action, new_data.id as old_id, new_data.dta as old_dta, new_data.id as new_id, new_data.dta as new_dta
25  from new_data where id not in (select id from ins_upd union all
26*                                select id from del_upd)
SQL> /
ACTION        OLD_ID O     NEW_ID N
Inserted                        4 Y
Updated            3 C          3 X
Deleted            2 B
No Change          1 A          1 A
SQL>