How can I generate a temporary table filled with dates in SQL Server 2000? - sql

I need to make a temporary table that holds of range of dates, as well as a couple of columns that hold placeholder values (0) for future use. The dates I need are the first day of each month between $startDate and $endDate where these variables can be several years apart.
My original sql statement looked like this:
select dbo.FirstOfMonth(InsertDate) as Month, 0 as Trials, 0 as Sales
into #dates
from customer
group by dbo.FirstOfMonth(InsertDate)
"FirstOfMonth" is a user-defined function I made that pretty much does what it says, returning the first day of the month for the provided date with the time at exactly midnight.
This produced almost exactly what I needed until I discovered there were occasionally gaps in my dates where I had a few months were there were no records insert dates. Since my result must still have the missing months I need a different approach.
I have added the following declarations to the stored procedure anticipating their need for the range of the dates I need ...
declare $startDate set $startDate = select min(InsertDate) from customer
declare $endDate set $endDate = select max(InsertDate) from customer
... but I have no idea what to do from here.
I know this question is similar to this question but, quite frankly, that answer is over my head (I don't often work with SQL and when I do it tends to be on older versions of SQL Server) and there are a few minor differences that are throwing me off.

I needed something similar, but all DAYS instead of all MONTHS.
Using the code from MatBailie as a starting point, here's the SQL for creating a permanent table with all dates from 2000-01-01 to 2099-12-31:
CREATE TABLE _Dates (
d DATE,
PRIMARY KEY (d)
)
DECLARE #dIncr DATE = '2000-01-01'
DECLARE #dEnd DATE = '2100-01-01'
WHILE ( #dIncr < #dEnd )
BEGIN
INSERT INTO _Dates (d) VALUES( #dIncr )
SELECT #dIncr = DATEADD(DAY, 1, #dIncr )
END

This will quickly populate a table with 170 years worth of dates.
CREATE TABLE CalendarMonths (
date DATETIME,
PRIMARY KEY (date)
)
DECLARE
#basedate DATETIME,
#offset INT
SELECT
#basedate = '01 Jan 2000',
#offset = 1
WHILE (#offset < 2048)
BEGIN
INSERT INTO CalendarMonths SELECT DATEADD(MONTH, #offset, date) FROM CalendarMonths
SELECT #offset = #offset + #offset
END
You can then use it by LEFT joining on to that table, for the range of dates you require.

I would probably use a Calendar table. Create a permanent table in your database and fill it with all of the dates. Even if you covered a 100 year range, the table would still only have ~36,525 rows in it.
CREATE TABLE dbo.Calendar (
calendar_date DATETIME NOT NULL,
is_weekend BIT NOT NULL,
is_holiday BIT NOT NULL,
CONSTRAINT PK_Calendar PRIMARY KEY CLUSTERED (calendar_date)
)
Once the table is created, just populate it once in a loop, so that it's always out there and available to you.
Your query then could be something like this:
SELECT
C.calendar_date,
0 AS trials,
0 AS sales
FROM
dbo.Calendar C
WHERE
C.calendar_date BETWEEN #start_date AND #end_date AND
DAY(C.calendar_date) = 1
You can join in the Customers table however you need to, outer joining on FirstOfMonth(InsertDate) = C.calendar_date if that's what you want.
You can also include a column for day_of_month if you want which would avoid the overhead of calling the DAY() function, but that's fairly trivial, so it probably doesn't matter one way or another.

This of course will not work in SQL-Server 2000 but in a modern database where you don't want to create a permanent table. You can use a table variable instead creating a table so you can left join the data try this. Change the DAY to HOUR etc to change the increment type.
declare #CalendarMonths table (date DATETIME, PRIMARY KEY (date)
)
DECLARE
#basedate DATETIME,
#offset INT
SELECT
#basedate = '01 Jan 2014',
#offset = 1
INSERT INTO #CalendarMonths SELECT #basedate
WHILE ( DATEADD(DAY, #offset, #basedate) < CURRENT_TIMESTAMP)
BEGIN
INSERT INTO #CalendarMonths SELECT DATEADD(HOUR, #offset, date) FROM #CalendarMonths where DATEADD(DAY, #offset, date) < CURRENT_TIMESTAMP
SELECT #offset = #offset + #offset
END

A starting point of a useful kludge to specify a range or specific list of dates:
SELECT *
FROM
(SELECT CONVERT(DateTime,'2017-1-1')+number AS [Date]
FROM master..spt_values WHERE type='P' AND number<370) AS DatesList
WHERE DatesList.Date IN ('2017-1-1','2017-4-14','2017-4-17','2017-12-25','2017-12-26')
You can get 0 to 2047 out of master..spt_values WHERE type='P', so that's five and a half year's worth of dates if you need it!

Tested below and it works, though it's a bit convoluted.
I assigned arbitrary values to the dates for the test.
DECLARE #SD smalldatetime,
#ED smalldatetime,
#FD smalldatetime,
#LD smalldatetime,
#Mct int,
#currct int = 0
SET #SD = '1/15/2011'
SET #ED = '2/02/2012'
SET #FD = (DATEADD(dd, -1*(Datepart(dd, #SD)-1), #sd))
SET #LD = (DATEADD(dd, -1*(Datepart(dd, #ED)-1), #ED))
SET #Mct = DATEDIFF(mm, #FD, #LD)
CREATE TABLE #MyTempTable (FoM smalldatetime, Trials int, Sales money)
WHILE #currct <= #Mct
BEGIN
INSERT INTO #MyTempTable (FoM, Trials, Sales)
VALUES
(DATEADD(MM, #currct, #FD), 0, 0)
SET #currct = #currct + 1
END
SELECT * FROM #MyTempTable
DROP TABLE #MyTempTable

For SQL Server 2000, this stackoverflow post looks promising for a way to temporarily generate dates calculated off of a start and end date. It's not exactly the same but quite similar. This post has a very in-depth answer on truncating dates, if needed.
In case anyone stumbles on this question and is working in PostgreSQL instead of SQL Server 2000, here is how you might do it there...
PostgreSQL has a nifty series generating function. For your example, you could use this series of all days instead of generating an entire calendar table, and then do groupings and matchups from there.
SELECT current_date + s.a AS dates FROM generate_series(0,14,7) AS s(a);
dates
------------
2004-02-05
2004-02-12
2004-02-19
(3 rows)
SELECT * FROM generate_series('2008-03-01 00:00'::timestamp,
'2008-03-04 12:00', '10 hours');
generate_series
---------------------
2008-03-01 00:00:00
2008-03-01 10:00:00
2008-03-01 20:00:00
2008-03-02 06:00:00
2008-03-02 16:00:00
2008-03-03 02:00:00
2008-03-03 12:00:00
2008-03-03 22:00:00
2008-03-04 08:00:00
(9 rows)
I would also look into date_trunc from PostgreSQL using 'month' for the truncator field to maybe refactor your original query to easily match with a date_trunc version of the calendar series.

select top (datediff(D,#start,#end)) dateadd(D,id-1,#start)
from BIG_TABLE_WITH_NO_JUMPS_IN_ID

declare #start datetime
set #start = '2016-09-01'
declare #end datetime
set #end = '2016-09-30'
create table #Date
(
table_id int identity(1,1) NOT NULL,
counterDate datetime NULL
);
insert into #Date select top (datediff(D,#start,#end)) NULL from SOME_TABLE
update #Date set counterDate = dateadd(D,table_id - 1, #start)
The code above should populate the table with all the dates between the start and end. You would then just join on this table to get all of the dates needed. If you only needed a certain day of each month, you could dateadd a month instead.

SELECT P.Id
, DATEADD ( DD, -P.Id, P.Date ) AS Date
FROM (SELECT TOP 1000 ROW_NUMBER () OVER (ORDER BY (SELECT NULL)) AS Id, CAST(GETDATE () AS DATE) AS Date FROM master.dbo.spt_values) AS P
This query returns a table calendar for the last 1000 days or so. It can be put in a temporary or other table.

Create a table variable containing a date for each month in a year:
declare #months table (reportMonth date, PRIMARY KEY (reportMonth));
declare #start date = '2018', #month int = 0; -- base 0 month
while (#month < 12)
begin
insert into #months select dateAdd(month, #month, #start);
select #month = #month + 1;
end
--verify
select * from #months;

This is by far the quickest method I have found (much quicker than inserting rows 1 by 1 in a WHILE loop):
DECLARE #startDate DATE = '1900-01-01'
DECLARE #endDate DATE = '2050-01-01'
SELECT DATEADD(DAY, sequenceNumber, #startDate) AS TheDate
INTO #TheDates
FROM (
SELECT ones.n + 10*tens.n + 100*hundreds.n + 1000*thousands.n + 10000*tenthousands.n AS sequenceNumber
FROM
(VALUES(0),(1),(2),(3),(4),(5),(6),(7),(8),(9)) ones(n),
(VALUES(0),(1),(2),(3),(4),(5),(6),(7),(8),(9)) tens(n),
(VALUES(0),(1),(2),(3),(4),(5),(6),(7),(8),(9)) hundreds(n),
(VALUES(0),(1),(2),(3),(4),(5),(6),(7),(8),(9)) thousands(n),
(VALUES(0),(1),(2),(3),(4),(5),(6),(7),(8),(9)) tenthousands(n)
WHERE ones.n + 10*tens.n + 100*hundreds.n + 1000*thousands.n + 10000*tenthousands.n <= DATEDIFF(day, #startDate, #endDate)
) theNumbers
SELECT *
FROM #TheDates
ORDER BY TheDate

The recursive answer:
DECLARE #startDate AS date = '20220315';
DECLARE #endDate AS date = '20230316'; -- inclusive
WITH cte_minutes(dt)
AS (
SELECT
DATEFROMPARTS(YEAR(#startDate), MONTH(#startDate), 1)
UNION ALL
SELECT
DATEADD(month, 1, dt)
FROM
cte_minutes
WHERE DATEADD(month, 1, dt) < #endDate
)
SELECT
dt
into #dates
FROM
cte_minutes
WHERE
dt >= #startDate
AND
dt <= #endDate
OPTION (MAXRECURSION 2000);
DROP TABLE dbo.#dates

Related

SQL Server - Efficient generation of dates in a range

Using SQL Server 2016.
I have a stored procedure that produces a list of options against a range of dates. Carriage options against days for clarity but unimportant to the specifics here.
The first step in the stored procedure generates a list of dates to store additional data against, and generating this list is taking substantially longer than the balance of the code. While this process is individual short, the number of calls means that this one piece of code is putting the system under more load than anything else.
With that in mind I have been testing efficiency of several options.
Iterative common table expression:
CREATE FUNCTION [dbo].[udf_DateRange_CTE] (#StartDate DATE,#EndDate DATE)
RETURNS #Return TABLE (Date DATE NOT NULL)
AS
BEGIN
WITH dates(date)
AS (SELECT #StartDate [Date]
UNION ALL
SELECT DATEADD(dd, 1, [Date])
FROM dates
WHERE [Date] < #EndDate
)
INSERT INTO #Return
SELECT date
FROM dates
OPTION (MAXRECURSION 0)
RETURN
END
A while loop:
CREATE FUNCTION [dbo].[udf_DateRange_While] (#StartDate DATE,#EndDate DATE)
RETURNS #Retun TABLE (Date DATE NOT NULL,PRIMARY KEY (Date))
AS
BEGIN
WHILE #StartDate <= #EndDate
BEGIN
INSERT INTO #Retun
VALUES (#StartDate)
SET #StartDate = DATEADD(DAY,1,#StartDate)
END
RETURN
END
A lookup from a pre-populated table of dates:
CREATE FUNCTION [dbo].[udf_DateRange_query] (#StartDate DATE,#EndDate DATE)
RETURNS #Return TABLE (Date DATE NOT NULL)
AS
BEGIN
INSERT INTO #Return
SELECT Date
FROM DateLookup
WHERE Date >= #StartDate
AND Date <= #EndDate
RETURN
END
In terms of efficiency I have test generating a years worth of dates, 1000 times and had the following results:
CTE: 10.0 Seconds
While: 7.7 Seconds
Query: 2.6 Seconds
From this the query is definitely the faster option but does require a permanent table of dates that needs to be created and maintained. This means that the query is no loner "self-contained" and it would be possible to request a date outside of the given date range.
Does anyone know of any more efficient ways of generating dates for a range, or any optimisation I can apply to the above?
Many thanks.
You can try like following. This should be fast compared CTE or WHILE loop.
DECLARE #StartDate DATETIME = Getdate() - 1000
DECLARE #EndTime DATETIME = Getdate()
SELECT *
FROM (SELECT #StartDate + RN AS DATE
FROM (SELECT ROW_NUMBER()
OVER (
ORDER BY (SELECT NULL)) RN
FROM master..[spt_values]) T) T1
WHERE T1.DATE <= #EndTime
ORDER BY DATE
Note: This will work for day difference <= 2537 days
If you want to support more range, you can use CROSS JOIN on master..[spt_values] to generate range between 0 - 6436369 days like following.
DECLARE #StartDate DATETIME = Getdate() - 10000
DECLARE #EndTime DATETIME = Getdate()
SELECT #StartDate + RN AS DATE FROM
(
SELECT ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) RN
FROM master..[spt_values] T1
CROSS JOIN master..[spt_values] T2
) T
WHERE RN <= DATEDIFF(DAY,#StartDate,#EndTime)

Add a record for every Week-Day of the year to a blank table

I have a blank table that has two columns [ID] and [MyDate].
I would like to populate that table with all of the days of the current year MINUS weekends.
Is there a way to do this with a SQL query?
In this case I am using MSSQL T-SQL
I do not have any example code, as I am at a loss on where to get started for this scenario.
Using a numbers (Tally) table helps you to avoid using loops.
If you don't already have a numbers table, you can use this script to create it:
SELECT TOP 10000 IDENTITY(int,0,1) AS Number
INTO Tally
FROM sys.objects s1
CROSS JOIN sys.objects s2
ALTER TABLE Tally ADD CONSTRAINT PK_NumbersTest PRIMARY KEY CLUSTERED (Number)
For more information about the creation of a numbers table, read this SO post.
Now that you have a numbers table, you can use a cte to generate the dates you want. I've used DATEFROMPARTS and GETDATE() to get Jauary 1st of the current year, if you are using a version of sql server below 2012 you need to use other methods for that:
DECLARE #StartDate Date,
#EndDate Date
SELECT #StartDate = DATEFROMPARTS(YEAR(GetDate()), 1, 1)
SELECT #EndDate = DATEADD(Year, 1, #StartDate)
Now, create a CTE to get the dates required using the numbers table, and insert the records from the cte to the table:
;WITH CTE AS
(
SELECT DATEADD(Day, Number, #StartDate) As TheDate
FROM Tally
WHERE DATEADD(Day, Number, #StartDate) < #EndDate
)
INSERT INTO WeekDays
SELECT TheDate
FROM CTE
WHERE DATEPART(WeekDay, TheDate) BETWEEN 2 AND 6
See a live demo on rextester.
This will do it. Here the 1 and the 7 represents Sunday and Saturday
CREATE TABLE T (
ID INT NOT NULL IDENTITY(1,1),
MyDate DATE NOT NULL)
DECLARE #Start DATE
DECLARE #End DATE
SET #Start = '20170101'
SET #End = '20171231'
WHILE #Start <= #End
BEGIN
IF (DATEPART(DW, #Start) NOT IN (1,7))
BEGIN
INSERT INTO T (MyDate) VALUES (#Start)
END
SET #Start = DATEADD(DAY, 1, #Start)
END
SELECT * FROM T
Here's my quick attempt at your problem. Just use your table instead
select
CAST('2017-03-15' as datetime) as datestuff
into #test
Delete from #test
DECLARE #Y datetime = CAST('2017-12-31' AS DATE)
while #y != '2017-01-01'
begin
if DATENAME(DW, #y) not IN ('SUNDAY', 'SATURDAY')
BEGIN
INSERT INTO #test
SELECT #y
END
SET #Y = DATEADD(DD, -1, #Y)
end
select * from #test

SQL Current status for given day (FOR loop)

For simplicity lets assume that I have a view with three fields
date_in (date)
Container (varchar)
date_out (date)
Now, the container is IN if the date_in is lesser or equal to given date and date_out is null or greater than given date. Now I am trying to count the containers for given time period. In pseudocode between two values STARTDATE and ENDDATE it would be something like
FOR X =STARTDATE, X<= ENDDADE, X++ {
if date_in <=X and date_out>x
count (container)
}
or closer to SQL:
declare #startdate date,
#d date;
set #startdate = '1/01/2014'
set #d = #startdate
"FOR on the #d variable would go here" {
select #d as SNAP_DATE, count (container) where date_in <#d
and (date_out is null or date_out> #d)
}
It might be simple - I guess I could make a new table and manually do multiple SELECT INTO (and later query from this new table) but its not very elegant solution.
Edit: just to precise - in the end I'd like to have something like:
DATE Count
1/02/2014 10
2/02/2014 15
...
7/03/2014 19
You could do this procedurally as follows:
Use a while loop to loop from start to end date.
Use a table variable to store each date-count pair.
Select from the table variable to get the summarised result.
declare #start date = '1/01/2014'
declare #end date = '7/03/2014'
declare #tbl table(Date date, Count int)
while(#start < #end)
begin
insert into #tbl
select #start, count(*)
from your_view
where (in_date < #start)
and ((out_date is null) or (out_date > #start))
set #start = dateadd(day, 1, #start)
end
select * from #tbl
You might be able to do something like the following. It uses a numbers table, which can be a real or derived table. It contains rows of integers. You need a table that begins with 0 and has enough values to cover your date range. Check here for more information on a numbers table.
DECLARE #StartDate DATE = '1/2/2014'
DECLARE #EndDate DATE = '1/4/2014'
SELECT DATEADD(d, n.num, #StartDate) AS DATE, COUNT(*) AS COUNT
FROM Numbers n
JOIN MyView mv ON mv.date_in < DATEADD(d, n.num, #StartDate)
AND (mv.date_out IS NULL OR mv.date_out > DATEADD(d, n.num, #StartDate))
WHERE DATEADD(d, n.num, #StartDate) BETWEEN #StartDate AND #EndDate
GROUP BY DATEADD(d, n.num, #StartDate)
ORDER BY DATEADD(d, n.num, #StartDate)
The numbers in the numbers table are converted to the list of dates between the date range. Each date is joined to your view based on the criteria you need.

SQL first of every month

Supposing that I wanted to write table valued function in SQL that returns a table with the first day of every month between the argument dates, what is the simplest way to do this?
For example fnFirstOfMonths('10/31/10', '2/17/11') would return a one-column table with 11/1/10, 12/1/10, 1/1/11, and 2/1/11 as the elements.
My first instinct is just to use a while loop and repeatedly insert first days of months until I get to before the start date. It seems like there should be a more elegant way to do this though.
Thanks for any help you can provide.
Something like this would work without being inside a function:
DECLARE #LowerDate DATE
SET #LowerDate = GETDATE()
DECLARE #UpperLimit DATE
SET #UpperLimit = '20111231'
;WITH Firsts AS
(
SELECT
DATEADD(DAY, -1 * DAY(#LowerDate) + 1, #LowerDate) AS 'FirstOfMonth'
UNION ALL
SELECT
DATEADD(MONTH, 1, f.FirstOfMonth) AS 'FirstOfMonth'
FROM
Firsts f
WHERE
DATEADD(MONTH, 1, f.FirstOfMonth) <= #UpperLimit
)
SELECT *
FROM Firsts
It uses a thing called CTE (Common Table Expression) - available in SQL Server 2005 and up and other database systems.
In this case, I start the recursive CTE by determining the first of the month for the #LowerDate date specified, and then I iterate adding one month to the previous first of month, until the upper limit is reached.
Or if you want to package it up in a stored function, you can do so, too:
CREATE FUNCTION dbo.GetFirstOfMonth(#LowerLimit DATE, #UpperLimit DATE)
RETURNS TABLE
AS
RETURN
WITH Firsts AS
(
SELECT
DATEADD(DAY, -1 * DAY(#LowerLimit) + 1, #LowerLimit) AS 'FirstOfMonth'
UNION ALL
SELECT
DATEADD(MONTH, 1, f.FirstOfMonth) AS 'FirstOfMonth'
FROM
Firsts f
WHERE
DATEADD(MONTH, 1, f.FirstOfMonth) <= #UpperLimit
)
SELECT * FROM Firsts
and then call it like this:
SELECT * FROM dbo.GetFirstOfMonth('20100522', '20100831')
to get an output like this:
FirstOfMonth
2010-05-01
2010-06-01
2010-07-01
2010-08-01
PS: by using the DATE datatype - which is present in SQL Server 2008 and newer - I fixed the two "bugs" that Richard commented about. If you're on SQL Server 2005, you'll have to use DATETIME instead - and deal with the fact you're getting a time portion, too.
create function dbo.fnFirstOfMonths(#d1 datetime, #d2 datetime)
returns table as return
select dateadd(m,datediff(m,0,#d1)+v.number,0) as FirstDay
from master..spt_values v
where v.type='P' and v.number between 0 and datediff(m, #d1, #d2)
and dateadd(m,datediff(m,0,#d1)+v.number,0) between #d1 and #d2
GO
Notes
master..spt_values is a source for general purpose sequence numbers in SQL Server
dateadd(m, datediff(m is a technique for working out the first day of month for any date
+v.number is used to increase it by one month each time
0 and datediff(m, #d1, #d2) this condition gives us all the numbers we need to generate a first-of-month date for each month between #d1 and #d2, inclusive of both months
and dateadd(m,datediff(m,0,#d1)+v.number,0) between #d1 and #d2 the final filter to verify that the first-of-month date generated is between #d1 and #d2
Performance comparison against marc_s's code
Summary
8220 ms (CTE)
4173 ms (master..spt_values)
Test
declare #t table (dt datetime)
declare #d datetime
declare #i int
set nocount on
set #d = GETDATE()
set #i = 0
while #i < 10000
begin
insert #t select * from dbo.getfirstofmonth('20090102', '20100506')
delete #t
set #i = #i + 1
end
print datediff(ms, #d, getdate())
set #d = GETDATE()
set #i = 0
while #i < 10000
begin
insert #t select * from dbo.fnfirstofmonths('20090102', '20100506')
delete #t
set #i = #i + 1
end
print datediff(ms, #d, getdate())
Performante
It will loop just between the months involved (4 times in the example):
set dateformat mdy;
declare #date1 smalldatetime,#date2 smalldatetime,#i int
set #date1= '10-31-2010'
set #date2= '02-17-2011'
set #i=1
while(#i<=DATEDIFF(mm,#date1,#date2))
begin
select convert(smalldatetime,CONVERT(varchar(6),DATEADD(mm,#i,#date1),112)+'01',112)
set #i=#i+1
end
I realize this isn't a function, but I'm going to throw this into the mix anyway.
select cal_date from calendar
where day_of_month = 1
and cal_date between '2011-01-01' and '2012-01-01'
This calendar table runs on a PostgreSQL server at work. I'll port it to SQL Server tonight, and run some speed comparisons. (Why? Because this stuff is fun, that's why.)
Just in case anybody is still reading this ...
I cannot imaging that any of the aforementioned functions is faster than this:
declare #DatFirst date = '20101031', #DatLast date = '21110217';
declare #DatFirstOfFirstMonth date = dateadd(day,1-day(#DatFirst),#DatFirst);
select DatFirstOfMonth = dateadd(month,n,#DatFirstOfFirstMonth)
from (
select top (datediff(month,#DatFirstOfFirstMonth,#DatLast)+1)
n=row_number() over (order by (select 1))-1
from (values (1),(1),(1),(1),(1),(1),(1),(1)) a (n)
cross join (values (1),(1),(1),(1),(1),(1),(1),(1)) b (n)
cross join (values (1),(1),(1),(1),(1),(1),(1),(1)) c (n)
cross join (values (1),(1),(1),(1),(1),(1),(1),(1)) d (n)
) x

Grouping by contiguous dates, ignoring weekends in SQL

I'm attempting to group contiguous date ranges to show the minimum and maximum date for each range. So far I've used a solution similar to this one: http://www.sqlservercentral.com/articles/T-SQL/71550/ however I'm on SQL 2000 so I had to make some changes. This is my procedure so far:
create table #tmp
(
date smalldatetime,
rownum int identity
)
insert into #tmp
select distinct date from testDates order by date
select
min(date) as dateRangeStart,
max(date) as dateRangeEnd,
count(*) as dates,
dateadd(dd,-1*rownum, date) as GroupID
from #tmp
group by dateadd(dd,-1*rownum, date)
drop table #tmp
It works exactly how I want except for one issue: weekends. My data sets have no records for weekend dates, which means any group found is at most 5 days. For instance, in the results below, I would like the last 3 groups to show up as a single record, with a dateRangeStart of 10/6 and a dateRangeEnd of 10/20:
Is there some way I can set this up to ignore a break in the date range if that break is just a weekend?
Thanks for the help.
EDITED
I didn't like my previous idea very much. Here's a better one, I think:
Based on the first and the last dates from the set of those to be grouped, prepare the list of all the intermediate weekend dates.
Insert the working dates together with weekend dates, ordered, so they would all be assigned rownum values according to their normal order.
Use your method of finding contiguous ranges with the following modifications:
1) when calculating dateRangeStart, if it's a weekend date, pick the nearest following weekday;
2) accordingly for dateRangeEnd, if it's a weekend date, pick the nearest preceding weekday;
3) when counting dates for the group, pick only weekdays.
Select from the resulting set only those rows where dates > 0, thus eliminating the groups formed only of the weekends.
And here's an implementation of the method, where it is assumed, that a week starts on Sunday (DATEPART returns 1) and weekend days are Sunday and Saturday:
DECLARE #tmp TABLE (date smalldatetime, rownum int IDENTITY);
DECLARE #weekends TABLE (date smalldatetime);
DECLARE #minDate smalldatetime, #maxDate smalldatetime, #date smalldatetime;
/* #1 */
SELECT #minDate = MIN(date), #maxDate = MAX(date)
FROM testDates;
SET #date = #minDate - DATEPART(dw, #minDate) + 7;
WHILE #date < #maxDate BEGIN
INSERT INTO #weekends
SELECT #date UNION ALL
SELECT #date + 1;
SET #date = #date + 7;
END;
/* #2 */
INSERT INTO #tmp
SELECT date FROM testDates
UNION
SELECT date FROM #weekends
ORDER BY date;
/* #3 & #4 */
SELECT *
FROM (
SELECT
MIN(date + CASE DATEPART(dw, date) WHEN 1 THEN 1 WHEN 7 THEN 2 ELSE 0 END)
AS dateRangeStart,
MAX(date - CASE DATEPART(dw, date) WHEN 1 THEN 2 WHEN 7 THEN 1 ELSE 0 END)
AS dateRangeEnd,
COUNT(CASE WHEN DATEPART(dw, date) NOT IN (1, 7) THEN date END) AS dates,
DATEADD(d, -rownum, date) AS GroupID
FROM #tmp
GROUP BY DATEADD(d, -rownum, date)
) s
WHERE dates > 0;