SQL cursor over categories? - sql-server

Question

I tried to post full SQL code to walk you through the data and conversions, but it wouldn't post here. Long story short, I end up with a data table like this:
Location Date Direction PreviousDirection Offset
site1 2013-07-22 11:30:45.000 302 302 0
site1 2013-07-22 11:31:45.000 322 302 20
site1 2013-07-22 11:32:45.000 9 322 47
site1 2013-07-22 11:33:45.000 9 9 0
site1 2013-07-22 11:34:45.000 0 9 -9
site2 2013-07-22 11:30:45.000 326 326 0
site2 2013-07-22 11:31:45.000 2 326 36
site2 2013-07-22 11:32:45.000 2 2 0
site2 2013-07-22 11:33:45.000 2 2 0
site2 2013-07-22 11:34:45.000 2 2 0
Location,Date is the primary key. I need help generating an [AdjustedDirection] column calculated as follows:
For first row (for each Location e.g. site1, site2): Since there is no previous row to calculate on, AdjustedDirection = first row's Direction.
After that, Second row AdjustedDirection: It's the first row's AdjustedDirection plus second row's offset.
Third row AdjustedDirection: It's the second row's AdjustedDirection plus third row's offset.
and so on...
I think this requires a cursor, but I don't know the syntax to do a cursor over multiple categories (Locations) and/or maybe there is a different answer. I can't describe how many steps and how complicated the process was to get to this step. I'm so close to the end and totally stuck here!
If anyone has a clue how to populate these AdjustedDirection values, please prove your awesomeness. Thanks!!
Results should look like this (date truncated for spacing, previous adjusted direction shown for clarity of how current row Adjusted is calculated):
Location Date Direction Offset PrevAdjDirection AdjustedDirection
site1 11:30:45.000 302 0 302 302
site1 11:31:45.000 322 20 302 322
site1 11:32:45.000 9 47 322 369
site1 11:33:45.000 9 0 369 369
site1 11:34:45.000 0 -9 369 360
site2 11:30:45.000 326 0 326 326
site2 11:31:45.000 2 36 326 362
site2 11:32:45.000 2 2 362 362
site2 11:33:45.000 2 2 362 362
site2 11:34:45.000 2 2 362 362
thanks!

Answer 1

Here is a solution using correlated subqueries, some of which can be replaced by window functions (the version of SQL Server makes a difference here).
You want to change your logic. Equivalent logic is:
For the first row, use the Direction
For subsequent rows, use the cumulative sum of the offsets excluding the first offset plus the direction from the first row.
The following calculates the appropriate variables using correlated subqueries, and then combines them using simple logic:
select t.*,
FirstOffset + coalesce(SumEarlierOffsets - FirstOffset + Offset, 0) as AdjustedOffset
from (select t.*,
(select Direction
from t t2
where t2.location = t.location
order by date asc
) as FirstDirection,
(select SUM(offset)
from t t2
where t2.location = t.location and
t2.date < t.date
) as SumEarlierOffsets,
(select Offset
from t t2
where t2.location = t.location
order by date asc
) as FirstOffset
from t
) t

Answer 2

I ended up dumping the current data into a temp table and doing a WHILE UPDATE like this
SELECT Location, Date, Direction, Offset, Adjusted = NULL
INTO #results
FROM t1
WHILE (
SELECT COUNT(*) FROM #results WHERE Adjusted IS NULL
) > 0
UPDATE TOP (1) t1
SET Adjusted = ISNULL(t2.Adjusted,ISNULL(t2.Direction,t1.Direction)) + t1.Offset
FROM #results t1
LEFT JOIN #results t2 ON t2.Location = t1.Location AND t2.Date = DateAdd(minute,-1,t1.Date)
WHERE t1.Adjusted IS NULL
Thanks for the input and inspiration!