Intermittent time series. Return first date of last contiguous block of most recent data in T-SQL

10/03/2011 11/03/2011 12/03/2011 13/03/2011 14/03/2011  15/03/2011  QUERYRESULT
a             a          a                      a            a       14/03/2011
               开发者_运维技巧                     b           b            b       13/03/2011
c                        c          c           c                    12/03/2011

So for the most recent block of contiguous dates, return the earliest date from that particular block, as shown in the 'queryresult' column.

I realise now that I've inadvertently transposed the data in the example I gave For source can assume schema is: [ID] 1,2,3,...; [type] a,b,c..; [date]

I can sort of see something with recursive CTE/window/ranking functions, but can't quite figure it out.

easier than I thought and doesn't need CTE or windowing/ranking functions:

assuming data is

ID  mydate  pfo
1   12/03/2011  a
4   13/03/2011  a
6   14/03/2011  a
10  16/03/2011  a
2   12/03/2011  b
5   13/03/2011  b
8   15/03/2011  b
3   12/03/2011  c
7   14/03/2011  c
9   15/03/2011  c
11  16/03/2011  c

then

SELECT 

    MAX( n.mydate)
   ,n.pfo

 FROM testtable n

 LEFT OUTER JOIN
 testtable n2 ON  n.pfo                  = n2.pfo 
              AND DATEADD(D,-1,n.mydate) = n2.mydate

 WHERE n2.mydate IS NULL

 GROUP BY n.pfo

;WITH data (type, date) AS (
  SELECT 'a', CAST('20110310' AS datetime) UNION ALL
  SELECT 'a', CAST('20110311' AS datetime) UNION ALL
  SELECT 'a', CAST('20110312' AS datetime) UNION ALL
  SELECT 'a', CAST('20110314' AS datetime) UNION ALL
  SELECT 'a', CAST('20110315' AS datetime) UNION ALL
  SELECT 'b', CAST('20110313' AS datetime) UNION ALL
  SELECT 'b', CAST('20110314' AS datetime) UNION ALL
  SELECT 'b', CAST('20110315' AS datetime) UNION ALL
  SELECT 'c', CAST('20110310' AS datetime) UNION ALL
  SELECT 'c', CAST('20110312' AS datetime) UNION ALL
  SELECT 'c', CAST('20110313' AS datetime) UNION ALL
  SELECT 'c', CAST('20110314' AS datetime)
),
grouped AS (
  SELECT
    type,
    date,
    groupID = DATEADD(day, -ROW_NUMBER() OVER (PARTITION BY type ORDER BY date), date)
  FROM data
),
startdates AS (
  SELECT
    type,
    groupStartDate = MIN(date)
  FROM grouped
  GROUP BY type, groupID
)
SELECT
  type,
  LastGroupStartDate = MAX(groupStartDate)
FROM startdates
GROUP BY type