Date range union

Question

Hi!
I have a subscription system, where people can subscribe to a publication during a certain period of time. I would like to know how long a person has subscribed to any publication, so i wonder does anyone know how to perform a "date range union"? By that, i mean that a query on this data:
with test_data as (select 1 id, to_date('01-JAN-2000','DD-MON-YYYY') start_date, to_date('27-JUN-2003','DD-MON-YYYY') stop_date from dual union
select 2, to_date('05-APR-2003','DD-MON-YYYY'), to_date('04-JAN-2004','DD-MON-YYYY') from dual union
select 3, to_date('03-JUL-2002','DD-MON-YYYY'), to_date('02-JUL-2008','DD-MON-YYYY') from dual)
select * from test_data;
should produce
START_DATE STOP_DATE
01-JAN-2000 00:00:00 02-JUL-2008 00:00:00
So there can be any number of periods, and they can overlap entirely as well as "on one end".
I have solved this using PL/SQL, but would like to know if there is a SQL solution. I have also solved this with a fixed number of periods (2,3...) but not with any number.

Answer 1

Sorry my fault. Soon after posting I found a case when the result was wrong, so I corrected the query (added default value 1 to lag(gap,1) i.e. lag(gap,1,1)) without leaving a notice. I'm home now - no database access - but I can comment to make it easier for you.
select the_start start_date,stop_date 
  from (
-- propagating the_start value: we keep data ordered by start date
--                              we carry the value of the_start from overlap row to the non-overlap row
        select start_date,stop_date,gap,              
               case when gap = 1 and lag(gap,1) over (order by start_date) = 0
                 then lag(the_start,1) over (order by start_date)
                 else start_date
               end the_start
          from (
-- excluding intermediate overlaps: gap = 0 and the_start is null
                select start_date,stop_date,gap,the_start                 
                  from (
-- identifying start days: we keep data ordered by start date
--                         when we have the first overlap (current gap = 0, but the gap on the previous row <> 0) we record the start_date as the_start
                        select start_date,stop_date,gap,
                               case when gap = 0 and lag(gap,1,1) over (order by start_date) = 1
                                    then start_date
                               end the_start
                          from (
-- identifying period breaks: we keep data ordered by start date
--                            when the start_date on the next row is not greater than the stop_date on the current row we have an overlap
                                select start_date,stop_date,
                                       case when lead(start_date,1) over (order by start_date) <= stop_date
                                            then 0
                                            else 1
                                       end gap
                                  from test_data
                 where gap = 1 or the_start is not null
where gap = 1
order by start_dateTo check intermediate results run step by step starting from the inner query (just comment out the outer queries). I'm pretty sure LAG, not LEAD
Regards
Etbin