Merge adjacent rows in SQL? - sql

Question

I'm doing some reporting based on the blocks of time employees work. In some cases, the data contains two separate records for what really is a single block of time.
Here's a basic version of the table and some sample records:
EmployeeID
StartTime
EndTime
Data:
EmpID Start End
----------------------------
#1001 10:00 AM 12:00 PM
#1001 4:00 PM 5:30 PM
#1001 5:30 PM 8:00 PM
In the example, the last two records are contiguous in time. I'd like to write a query that combines any adjacent records so the result set is this:
EmpID Start End
----------------------------
#1001 10:00 AM 12:00 PM
#1001 4:00 PM 8:00 PM
Ideally, it should also be able to handle more than 2 adjacent records, but that is not required.

Answer 1

This article provides quite a few possible solutions to your question
http://www.sqlmag.com/blog/puzzled-by-t-sql-blog-15/tsql/solutions-to-packing-date-and-time-intervals-puzzle-136851
This one seems like the most straight forward:
WITH StartTimes AS
(
SELECT DISTINCT username, starttime
FROM dbo.Sessions AS S1
WHERE NOT EXISTS
(SELECT * FROM dbo.Sessions AS S2
WHERE S2.username = S1.username
AND S2.starttime < S1.starttime
AND S2.endtime >= S1.starttime)
),
EndTimes AS
(
SELECT DISTINCT username, endtime
FROM dbo.Sessions AS S1
WHERE NOT EXISTS
(SELECT * FROM dbo.Sessions AS S2
WHERE S2.username = S1.username
AND S2.endtime > S1.endtime
AND S2.starttime <= S1.endtime)
)
SELECT username, starttime,
(SELECT MIN(endtime) FROM EndTimes AS E
WHERE E.username = S.username
AND endtime >= starttime) AS endtime
FROM StartTimes AS S;

Answer 2

If this is strictly about adjacent rows (not overlapping ones), you could try the following method:
Unpivot the timestamps.
Leave only those that have no duplicates.
Pivot the remaining ones back, coupling every Start with the directly following End.
Or, in Transact-SQL, something like this:
WITH unpivoted AS (
SELECT
EmpID,
event,
dtime,
count = COUNT(*) OVER (PARTITION BY EmpID, dtime)
FROM atable
UNPIVOT (
dtime FOR event IN (StartTime, EndTime)
) u
)
, filtered AS (
SELECT
EmpID,
event,
dtime,
rowno = ROW_NUMBER() OVER (PARTITION BY EmpID, event ORDER BY dtime)
FROM unpivoted
WHERE count = 1
)
, pivoted AS (
SELECT
EmpID,
StartTime,
EndTime
FROM filtered
PIVOT (
MAX(dtime) FOR event IN (StartTime, EndTime)
) p
)
SELECT *
FROM pivoted
;
There's a demo for this query at SQL Fiddle.

Answer 3

CTE with cumulative sum:
DECLARE #t TABLE(EmpId INT, Start TIME, Finish TIME)
INSERT INTO #t (EmpId, Start, Finish)
VALUES
(1001, '10:00 AM', '12:00 PM'),
(1001, '4:00 PM', '5:30 PM'),
(1001, '5:30 PM', '8:00 PM')
;WITH rowind AS (
SELECT EmpId, Start, Finish,
-- IIF returns 1 for each row that should generate a new row in the final result
IIF(Start = LAG(Finish, 1) OVER(PARTITION BY EmpId ORDER BY Start), 0, 1) newrow
FROM #t),
groups AS (
SELECT EmpId, Start, Finish,
-- Cumulative sum
SUM(newrow) OVER(PARTITION BY EmpId ORDER BY Start) csum
FROM rowind)
SELECT
EmpId,
MIN(Start) Start,
MAX(Finish) Finish
FROM groups
GROUP BY EmpId, csum

Answer 4

I have changed a lil' bit the names and types to make the example smaller but this works and should be very fast and it has no number of records limit:
with cte as (
select
x1.id
,x1.t1
,x1.t2
,case when x2.t1 is null then 1 else 0 end as bef
,case when x3.t1 is null then 1 else 0 end as aft
from x x1
left join x x2 on x1.id=x2.id and x1.t1=x2.t2
left join x x3 on x1.id=x3.id and x1.t2=x3.t1
where x2.id is null
or x3.id is null
)
select
cteo.id
,cteo.t1
,isnull(z.t2,cteo.t2) as t2
from cte cteo
outer apply (select top 1 *
from cte ctei
where cteo.id=ctei.id and cteo.aft=0 and ctei.t1>cteo.t1
order by t1) z
where cteo.bef=1
and the fiddle for it : http://sqlfiddle.com/#!3/ad737/12/0

Answer 5

Option with Inline User-Defined Function AND CTE
CREATE FUNCTION dbo.Overlap
(
#availStart datetime,
#availEnd datetime,
#availStart2 datetime,
#availEnd2 datetime
)
RETURNS TABLE
RETURN
SELECT CASE WHEN #availStart > #availEnd2 OR #availEnd < #availStart2
THEN #availStart ELSE
CASE WHEN #availStart > #availStart2 THEN #availStart2 ELSE #availStart END
END AS availStart,
CASE WHEN #availStart > #availEnd2 OR #availEnd < #availStart2
THEN #availEnd ELSE
CASE WHEN #availEnd > #availEnd2 THEN #availEnd ELSE #availEnd2 END
END AS availEnd
;WITH cte AS
(
SELECT EmpID, Start, [End], ROW_NUMBER() OVER (PARTITION BY EmpID ORDER BY Start) AS Id
FROM dbo.TableName
), cte2 AS
(
SELECT Id, EmpID, Start, [End]
FROM cte
WHERE Id = 1
UNION ALL
SELECT c.Id, c.EmpID, o.availStart, o.availEnd
FROM cte c JOIN cte2 ct ON c.Id = ct.Id + 1
CROSS APPLY dbo.Overlap(c.Start, c.[End], ct.Start, ct.[End]) AS o
)
SELECT EmpID, Start, MAX([End])
FROM cte2
GROUP BY EmpID, Start
Demo on SQLFiddle