This looks like a Packing Intervals problem. See the post by Itzik Ben-Gan for all the details and what indexes he recommends to make it work efficiently. He presents a solution without recursive CTE.
Two notes.
The query below assumes that intervals are [closed; open), i.e. StartDate is inclusive and EndDate is exclusive. This way to represent such data is often the most convenient. (in the same sense as having arrays as zero-based instead of 1-based is usually more convenient in programming languages).
I added a RowID column to have unambiguous sorting.
Sample data
DECLARE @T TABLE
(
RowID int IDENTITY,
id int,
StartDate date,
EndDate date,
tp int
);
INSERT INTO @T(Id, StartDate, EndDate, tp) VALUES
(1, '2012-02-18', '2012-03-18', 1),
(1, '2012-03-17', '2012-06-29', 1),
(1, '2012-06-27', '2012-09-27', 1),
(1, '2014-08-23', '2014-09-24', 3),
(1, '2014-09-23', '2014-10-24', 3),
(1, '2014-10-23', '2014-11-24', 3),
(2, '2015-07-04', '2015-08-06', 1),
(2, '2015-08-04', '2015-09-06', 1),
(3, '2013-11-01', '2013-12-01', 0),
(3, '2018-01-09', '2018-02-09', 0);
-- Make EndDate an opened interval, make it exclusive
-- [Start; End)
UPDATE @T
SET EndDate = DATEADD(day, 1, EndDate)
;
Recommended indexes
-- indexes to support solutions
CREATE UNIQUE INDEX idx_start_id ON T(id, tp, StartDate, RowID);
CREATE UNIQUE INDEX idx_end_id ON T(id, tp, EndDate, RowID);
Query
Read the Itzik's post to understand what is going on. He has nice illustrations there. In short, each timestamp (start or end) is treated as an event. Each event has a + or - type. Each time we encounter a + event (some interval starts) we increase the running counter. Each time we encounter a - event (some interval ends) we decrease the running counter. When the running counter is 0 it means that the streak of overlapping intervals is over.
I took Itzik's query as is and simply changed the column names to match your names.
WITH C1 AS
-- let e = end ordinals, let s = start ordinals
(
SELECT
RowID, id, tp, StartDate AS ts, +1 AS EventType,
NULL AS e,
ROW_NUMBER() OVER(PARTITION BY id, tp ORDER BY StartDate, RowID) AS s
FROM @T
UNION ALL
SELECT
RowID, id, tp, EndDate AS ts, -1 AS EventType,
ROW_NUMBER() OVER(PARTITION BY id, tp ORDER BY EndDate, RowID) AS e,
NULL AS s
FROM @T
),
C2 AS
-- let se = start or end ordinal, namely, how many events (start or end) happened so far
(
SELECT C1.*,
ROW_NUMBER() OVER(PARTITION BY id, tp ORDER BY ts, EventType DESC, RowID) AS se
FROM C1
),
C3 AS
-- For start events, the expression s - (se - s) - 1 represents how many sessions were active
-- just before the current (hence - 1)
--
-- For end events, the expression (se - e) - e represents how many sessions are active
-- right after this one
--
-- The above two expressions are 0 exactly when a group of packed intervals
-- either starts or ends, respectively
--
-- After filtering only events when a group of packed intervals either starts or ends,
-- group each pair of adjacent start/end events
(
SELECT id, tp, ts,
((ROW_NUMBER() OVER(PARTITION BY id, tp ORDER BY ts) - 1) / 2 + 1)
AS grpnum
FROM C2
WHERE COALESCE(s - (se - s) - 1, (se - e) - e) = 0
)
SELECT id, tp, MIN(ts) AS StartDate, DATEADD(day, -1, MAX(ts)) AS EndDate
FROM C3
GROUP BY id, tp, grpnum
ORDER BY id, tp, StartDate;
Result
+----+----+------------+------------+
| id | tp | StartDate | EndDate |
+----+----+------------+------------+
| 1 | 1 | 2012-02-18 | 2012-09-27 |
| 1 | 3 | 2014-08-23 | 2014-11-24 |
| 2 | 1 | 2015-07-04 | 2015-09-06 |
| 3 | 0 | 2013-11-01 | 2013-12-01 |
| 3 | 0 | 2018-01-09 | 2018-02-09 |
+----+----+------------+------------+
