I've a table, which describes work slices of a business working calendar: (date format is 24 hours format)
PK | STARTDATE | ENDDATE
__________________________________________
1 | 2012/07/21 02:00 | 2012/07/21 04:00
2 | 2012/07/21 03:00 | 2012/07/21 10:00
3 | 2012/07/21 06:00 | 2012/07/21 17:00
4 | 2012/07/21 18:00 | 2012/07/21 19:00
Now, I like to merge the date ranges (within a given start and end date) like this:
PK | STARTDATE | ENDDATE
__________________________________________
1 | 2012/07/21 02:00 | 2012/07/21 17:00
2 | 2012/07/21 18:00 | 2012/07/21 19:00
Is there a way to do this with SQL97 standard? If so, what is with other operations (e.g. if I want to to an invered merge, the result should be
PK | STARTDATE | ENDDATE
__________________________________________
1 | 2012/07/21 00:00 | 2012/07/21 02:00
2 | 2012/07/21 19:00 | 2012/07/22 00:00
Here's an example using SQL Server syntax. First it determines the "heads", or rows that have no previous overlapping rows. To determine the last "child" of a "head", it looks for the last row that is smaller than the next "head". Here's the SQL:
; with heads as
(
select row_number() over (order by head.StartDate) as PK
, *
from YourTable head
where not exists
(
select *
from YourTable prev
where prev.StartDate < head.StartDate
and head.StartDate < prev.EndDate
)
)
select row_number() over (order by h.StartDate) as PK
, h.StartDate
, max(yt.EndDate) as EndDate
from heads h
left join
heads nh
on nh.PK = h.PK + 1
left join
YourTable yt
on h.StartDate <= yt.StartDate
and (yt.StartDate < nh.StartDate or nh.StartDate is null)
group by
h.StartDate