SQL split eggs in two baskets - date alternation

Question

I have an interesting and specific problem. Specific, because I cannot explain it simply, but I think it should be reasonably easy to solve. With that said, I cannot find similar question, because I don’t know how to look for it!

In one table, among other columns, there is a date column, with same multiple entries repeating for a year or so in a specific pattern.

For example:

TermDate

• 2013/07/16
• 2013/07/16
• 2013/07/16
• 2013/07/23
• 2013/07/23
• 2013/07/23
• 2013/07/23
• 2013/07/30
• 2013/07/30
• 2013/07/30
• 2013/07/30 …

(Every Tuesday)

I need to change those dates to Mondays and Wednesdays, so that ‘2013/07/16’ would become ‘2013/07/15’, ‘2013/07/23’ would then be ‘‘2013/07/17’’, and ‘2013/07/30’ would be next first available Monday – ‘2013/07/22’, etc.

Do you see what I mean? Obviously I would need CASE statement for this, but does anyone have an idea how exactly to write it?

Thanks!

I am using the MS SQL SERVER by the way!

Answer 1

SQL to round a date down to the start of the week...

DATEADD(WEEK, DATEDIFF(WEEK, 0, <theDate>), 0)

NOTE: This is NOT affected by your DATEFIRST environment setting.

SQL to conditionally round it down to a Monday or Wednesday...

CASE WHEN DATEPART(DW, <theDate>) IN (3, 4, 5, 6, 7) THEN
  DATEADD(WEEK, DATEDIFF(WEEK, 0, <theDate>), 0)
ELSE
  DATEADD(WEEK, DATEDIFF(WEEK, 0, <theDate>), 0) + 2
END

NOTE : DATEPART() _IS_ affected by your DATEFIRST environment setting.
       This code has assumed SET DATEFIRST 7

Then you can put all of your records into a sequence. If it's position 1 you add two days to the base date (Monday => Wednesday), if it's position 2 you add seven days (Monday => Next Monday), position three adds nine days, etc.

This simplifies down to (position / 2) * 7 + (position % 2) * 2...

position : (position / 2) * 7 + (position % 2) * 2
--------------------------------------------------
       0 : 0
       1 : 2
       2 : 7
       3 : 9
       4 : 14
       5 : 16
       6 : 21

With a note that if your base date happens to be a Wednesday, you need to treat all of your other dates as having a position one higher...

So, I'd do this...

SELECT
  term_lookup.TermDate,
  base_lookup.base_date
  + ((term_lookup.term_id + base_lookup.base_id) / 2) * 7
  + ((term_lookup.term_id + base_lookup.base_id) % 2) * 2   AS new_date
FROM
(
  SELECT
    DATEADD(WEEK, DATEDIFF(WEEK, 0, MIN(TermDate)), 0)                  AS base_date,
    CASE WHEN DATEPART(DW, MIN(TermDate)) IN (1, 2) THEN 0 ELSE 1 END   AS base_id
  FROM
    yourTable
)
  AS base_lookup
CROSS JOIN
(
  SELECT
    TermDate,
    DENSE_RANK() OVER (ORDER BY TermDate) - 1   AS term_id
  FROM
    yourTable
  GROUP BY
    TermDate
)
  AS term_lookup

If that gives you the right results, join them back on to your data to do the update.

Answer 2

try this

UPDATE Table1
SET TermDate = DATEADD(DAY, CASE DATEPART(weekday, TermDate ) 
                        WHEN 1 THEN 1
                        WHEN 2 THEN 0
                        WHEN 3 THEN 1
                        WHEN 4 THEN 0
                        WHEN 5 THEN 4
                        WHEN 6 THEN 3
                        WHEN 7 THEN 2
                     END, TermDate )