SQL Query on Date Range

Question

Hi, I am using Oracle 10g. I want a query which gives me below output.
Data Setup
========
create table t_date_range (
ID     number (2),
start_date     date,
end_date     date);
insert into t_date_range values (1,to_date('20110101', 'YYYYMMDD'),to_date('20110331', 'YYYYMMDD'));
insert into t_date_range values (2,to_date('20110401', 'YYYYMMDD'),to_date('20110531', 'YYYYMMDD'));
insert into t_date_range values (3,to_date('20110701', 'YYYYMMDD'),to_date('20110731', 'YYYYMMDD'));
insert into t_date_range values (4,to_date('20110901', 'YYYYMMDD'),to_date('20111130', 'YYYYMMDD'));
insert into t_date_range values (5,to_date('20111201', 'YYYYMMDD'),to_date('20111231', 'YYYYMMDD'));
commit;
SQL> select ID, to_char(start_date,'DD-MON-YYYY') START_DATE, to_char(end_date,'DD-MON-YYYY') END_DATE from t_date_range;
ID START_DATE END_DATE
1 01-JAN-2011 31-MAR-2011
2 01-APR-2011 31-MAY-2011
3 01-JUL-2011 31-JUL-2011
4 01-SEP-2011 25-OCT-2011
5 26-OCT-2011 30-NOV-2011
6 01-DEC-2011 31-DEC-2011
6 rows selected.
I want result in this form:
START_DATE END_DATE
01-JAN-2011 31-MAY-2011
01-JUL-2011 31-JUL-2011
01-SEP-2011 31-DEC-2011
Means if there is a difference of exact one day between "start_date of 2nd row" and "end_date of first row" then make a single row as shows in above results set.
Thanks!

Answer 1

Hi,
Solomon Yakobson wrote:
Keep in mind, Franks's solution assumes date ranges do not overlap. ...That's true. If rows can overlap, then we might need to use two sub-queries, like you did: one to see if a new group starts with this row, and another to count how many new groups have already started.
The solution you posted assumes a relationship between id and dates. If we add a row like this to the sample data:
insert into t_date_range values (6,to_date('20101201', 'YYYYMMDD'),to_date('20121231', 'YYYYMMDD'));that overlaps all the others, then how would that solution work?
LAG won't work, because no matter how we sort the rows, we can't be sure that overlapping rows will be consecutive.
Here's one way to handle overlapping rows:
WITH     got_new_grp          AS
     SELECT     id, start_date, end_date
     ,     CASE
              WHEN  start_date > 1 + MAX (end_date) OVER ( ORDER BY  start_date
                                                                    ,             id
                                                                    ROWS BETWEEN  UNBOUNDED PRECEDING
                                              AND      1      PRECEDING
              THEN  1
          END     AS new_grp
     FROM     t_date_range
,     got_grp          AS
     SELECT  start_date, end_date
     ,     COUNT (new_grp) OVER ( ORDER BY  start_date
                                 ,           id
                         ) AS grp
     FROM    got_new_grp
SELECT       MIN (start_date)     AS start_date
,       MAX (end_date)     AS end_date
FROM       got_grp
GROUP BY  grp
ORDER BY  start_date
;This solution uses id (assumed to be unique) just to get a consistent ordering, in case 2 (or more) rows have exactly the same start_date. It does not assume any relationship between ids and either start_date or end_date.
I'm also assuming that start_date<=end_date on each row.
Edited by: Frank Kulash on May 11, 2011 12:54 PM
The above refers to the solution you oriognally posted, which was:
with t1 as (
            select  id,
                    start_date,
                    end_date,
                    case
                      when lag(end_date) over(order by id) + 1 >= start_date then 0
                      else 1
                    end start_of_group
              from  t_date_range
     t2 as (
            select  id,
                    start_date,
                    end_date,
                    sum(start_of_group) over(order by id) grp
              from  t1
select  min(start_date) start_date,
        max(end_date) end_date
  from  t2
  group by grp
  order by grp
;