Number of weeks and partial weeks between two days calculated wrong - sql-server

Why is this returning 4 instead of 6?
SET DATEFIRST 1
SELECT DATEDIFF(WEEK, CAST('2017-01-01' AS DATE), CAST('2017-01-31' AS DATE))
Is week as datepart calculating only weeks from monday - sunday (whole weeks)? How to get all weeks - including those which doesn't have seven days ? In the case above answer should be 6.

DATEDIFF counts transitions, not periods (e.g. look at DATEDIFF(year,'20161231','20170101')). It also treats Sunday as the first day of the week. So, how do we compensate for these features? First, we shift our dates so that Mondays are the new Sundays, and second we add 1 to compensate for the Fence-Post error:
declare #Samples table (
StartAt date not null,
EndAt date not null,
SampleName varchar(93) not null
)
insert into #Samples (StartAt,EndAt,SampleName) values
('20170101','20170131','Question - 6'),
('20170102','20170129','Exactly 4'),
('20170102','20170125','3 and a bit, round to 4'),
('20170101','20170129','4 and 1 day, round to 5')
--DATEDIFF counts *transitions*, and always considers Sunday the first day of the week
--We subtract a day from each date so that we're effectively treating Monday as the first day of the week
--We also add one because DATEDIFF counts transitions but we want periods (FencePost/FencePanel)
select *,
DATEDIFF(WEEK, DATEADD(day,-1,StartAt), DATEADD(day,-1,EndAt)) +1
as NumWeeks
from #Samples
Results:
StartAt EndAt SampleName NumWeeks
---------- ---------- -------------------------- -----------
2017-01-01 2017-01-31 Question - 6 6
2017-01-02 2017-01-29 Exactly 4 4
2017-01-02 2017-01-25 3 and a bit, round to 4 4
2017-01-01 2017-01-29 4 and 1 day, round to 5 5
If this doesn't match what you want, perhaps you can adopt and adapt my #Samples table to show the results you do expect.

What you ask though, is how many weeks are covered by a range, not how many weeks are between two dates.
DATEDIFF always uses Sunday when calculating week transitions.
This isn't a bug, it's done to ensure the function is deterministic and returns the same value, for every query, no matter the DATEFIRST setting. From the documentation
Specifying SET DATEFIRST has no effect on DATEDIFF. DATEDIFF always uses Sunday as the first day of the week to ensure the function is deterministic.
One solution would be to calculate the difference between the week numbers of the start and end dates, when the first day is Monday. 1 is added to the difference to take account of the first week as well:
SET DATEFIRST 1;
select 1 +datepart(WEEK,'20170131') - datepart(WEEK,'20170101')
That's a fragile calculation though that breaks if DATEFIRST changes or if one of the dates is on a different year.
You could use ISO Weeks to get rid of SET DATEFIRST:
select 1 +datepart(ISO_WEEK,'20170131') - datepart(ISO_WEEK,'20170101')
but that would fail for 2017-01-01 because Sunday is counted as Week 52 of the previous year.
A far better solution though would be to count the distinct week numbers using a Calendar table that contains dates and different week numbers to cover multiple business requirements, eg both normal and ISO Week numbers, or business calendars based on a 4-4-5 calendar.
In this case, you could just count distinct week numbers:
SELECT COUNT(DISTINCT Calendar.IsoWeek )
from Calendar
where date between '20170101' and '20170131'
If the table doesn't have an ISO Week column, you can use DATEPART:
select count (distinct datepart(ISO_WEEK,date) )
from Calendar
where date between '20170101' and '20170131'

Related

##DATEFIRST not returning results as described in the docs

I have read this documentation
So I tried this experiment
declare #t table (test date)
insert into #t values ('20220404'), ('20220405'),('20220406'),('20220407'),('20220408'),('20220409'),('20220410')
select datename(weekday, test),
datepart(weekday, test)
from #t
it returns this
COLUMN1
COLUMN2
Monday
2
Tuesday
3
Wednesday
4
Thursday
5
Friday
6
Saturday
7
Sunday
1
I checked my value or ##DATEFIRST
select ##DATEFIRST
it returns 7
So why do I not get this result then as described in the docs?
COLUMN1
COLUMN2
Monday
1
Tuesday
2
Wednesday
3
Thursday
4
Friday
5
Saturday
6
Sunday
7
EDIT
this is what I see in the docs
I think you may be misunderstanding the docs. The docs for DATEFIRST say, as you've seen:
Sets the first day of the week
So, the value of DATEFIRST determines which day gets numbered 1, the first day of the week. With DATEFIRST set to 7, as the table goes on to show, Sunday will be considered the first day of the week - day number 1.
With that setting, DATEPART for weekday will return 1 for any Sunday, because Sunday is considered the first day of the week.
It is perhaps unfortunate that numbers are used as the argument to SET DATEFIRST, since naturally this confusion arises. It might have been nice if we could say SET DATEFIRST Sunday to make it obvious what we mean, but unfortunately that's not the syntax.

Return dates from last/this week depending on the current time and when "production" started

Sorry if the Title is confusing but it's hard to explain what I'm after in one phrase.
I'm currently producing a report based on the production for the week. I start off my CTE construction with the following to get the days Monday to Friday of the current week:
WITH
cte_Date AS
(
SELECT
CAST(DateTime AS date) AS Date
FROM
( VALUES
(GETDATE()
)
, (DATEADD(day,-1,GETDATE()))
, (DATEADD(day,-2,GETDATE()))
, (DATEADD(day,-3,GETDATE()))
, (DATEADD(day,-4,GETDATE()))
, (DATEADD(day,-5,GETDATE()))
, (DATEADD(day,-6,GETDATE())) ) AS LastSevenDays(DateTime)
WHERE
DATENAME(weekday, DateTime) = 'Monday'
UNION ALL
SELECT
DATEADD(day,1,Date)
FROM
cte_Date
WHERE
DATENAME(weekday,Date) <> 'Friday'
)
This is working fine. I have made the report available to users so they can run it anytime however sometimes nobody is available to run it last thing Friday. This means they don't get to see the full production for Friday and then the following week the CTE days change.
I'm trying to keep this a one-click affair so rather than introduce date parameters I proposed to the users that we adjust the query such that if they run the report before midday on "Monday" then it will show them last week's figures and they were happy with this (me and my big mouth). I put Monday in quotes because what we really mean of course is the first production day of the week.
My primary data table (which we'll call MyData) has a datetime field named DateTime (really!) that I can reference to determine the first day of production for the week.
One final caveat: Due to the layout of the report the users insisted that they always want to see the five days Monday to Friday, even if there is no production on a given day. (Consequently I do a LEFT JOIN from cte_Date to all other tables required.) So to be clear, right now as I'm typing this it's 11:45am local time on Tuesday and yesterday happened to be a public holiday here so running the report now should return Monday to Friday last week, but running it in 20 minutes time should return Monday to Friday this week.
Please help, my poor brain is getting twisted trying to figure it out.
There are a few different ways you can tackle this, but they all boil down to the same thing: you need a way of figuring out whether it's before or after 12pm on the first working day of the current week, then you need to get the Monday of the current "production week".
Let's just say, for simplicity's sake, you have some sort of table that contains public holidays (or non-production days). To find out whether it's the first day of the current production week, you basically just have to add the number of days in a row since the start of the week that have been public holidays.
Then you need to figure out whether it's before or after 12pm of that day.
If it's before you want last week's Monday-Friday. If it's after, you want this week's Monday-Friday.
Here's one way you might do this:
DECLARE #NonProductionDays TABLE (NPD DATE UNIQUE NOT NULL); -- Public holiday table.
INSERT #NonProductionDays (NPD) VALUES ('2017-09-25');
DECLARE #i INT = -- You don't need a variable for this, but just to keep things simple...
(
SELECT COUNT(*) -- Extract number of public holidays in a row this week before current date.
FROM #NonProductionDays AS N
WHERE DATEDIFF(WEEK, 0, N.NPD) = DATEDIFF(WEEK, 0, GETDATE())
AND N.NPD <= GETDATE()
AND (DATENAME(WEEKDAY, N.NPD) = 'Monday' OR EXISTS (SELECT 1 FROM #NonProductionDays AS N2 WHERE N2.NPD = DATEADD(DAY, -1, N.NPD)))
);
SELECT D = CAST(DATEADD(DAY, T.N, DATEADD(WEEK, DATEDIFF(HOUR, DATEADD(DAY, #i, '1900-01-01 12:00:00'), GETDATE()) / 24 / 7, '1900-01-01')) AS DATE)
FROM (VALUES (0), (1), (2), (3), (4)) AS T(N);
/*
Breaking this down:
X = DATEADD(DAY, #i, '1900-01-01 12:00:00')
-- Adds the number of NPD days this week to '1900-01-01 12:00:00'
-- So, for example, X would be '1900-01-02 12:00:00' this week
Y = DATEDIFF(HOUR, X, GETDATE()) / 24 / 7
-- The number of weeks between X and now, by taking the number of hours and dividing by 24 then by 7
-- The division is necessary to compare the hour.
-- So, for example, as of 11am on the September 26 2017, you'd get 6142.
-- As of 12pm on September 26 2017, you'd get 6143.
Z = DATEADD(WEEK, Y, '1900-01-01')
-- Just adds Y weeks to 1900-01-01, which was a Monday. This tells you the Monday of the current "production week".
-- So, for example, as of 11am on September 26 2017, you'd get '2017-09-18 00:00:00.000'.
-- As of 12pm on September 26 2017, you'd get '2017-09-25 00:00:00.000'.
Then we cast this as a date and add 0/1/2/3/4 days to it to get Monday, Tuesday, Wednesday, Thursday and Friday of the current "production week".
*/
I'm not sure I came up with the most efficient approach, but after a week of tossing it about in my brain this is what I came up with. I approached the problem from the opposite direction of that suggested by #ZLK.
My existing logic was already giving me the Monday of this week so in a subquery I looked for the first production record after Monday, stripped off the time with a DATEDIFF and made it midday with a DATEADD. I was then able to compare the current Date/Time with midday of the first production day to determine whether to reduce the date by one week.
I replaced this SELECT clause:
SELECT
CAST(DateTime AS date) AS Date
with this one:
SELECT -- Monday this week if it's after midday on the first production day otherwise Monday last week
DATEADD(week,IIF(GETDATE()>=DATEADD(hour,12,(
SELECT DATEDIFF(day,0,MIN(DateTime))
FROM MyData
WHERE CAST(MyData.DateTime AS date) >= CAST(LastSevenDays.DateTime AS date)
)),0,-1),CAST(LastSevenDays.DateTime AS date)) AS Date
To cater for the case where a new week has commenced but the operator runs the report before production starts I carefully arranged the boolean condition inside my IIF clause so that the empty result set from the subquery would mean the test returned FALSE and the operator would still see last week's figures.
(#ZLK, Thanks for your input - you did help my thinking a bit but I don't think your answer should be marked as correct. What I've come up with here is what I was originally requesting and didn't require the use of a static table.)

Storing a trial duration in years, months, days, hours in SQL Server

UPDATE: Just to be clear for future viewers, the accepted answer does not propose a different or better approach to those given in this question (it uses a "total number of days" (or hours) column). From the discussion in the comments, after the initial confusion, it resulted that there's likely no way to cover both ease of use and precision, so using a single but inexact column in "days" was the approach we went with, and the accepted answer represents that. Cheers!
Our products have licenses, and those licenses might be trials, having an end time starting from the point when they are first used. The trial period consists of a certain number of years, months, days and, ideally, hours as well.
We wanted to store this period in an SQL Server database, but are struggling. There doesn't seem to be a data type for this, based on this documentation.
DateTimeOffset is not a real "offset" but a date AND its offset from UTC (please correct me if I'm wrong), so it doesn't seem to support concepts like "0 months".
Storing a "total number of seconds" or "total number of hours" is susceptible to time changes around the globe. It is also hard to look at a value (as a human) and have a notion of how long the trial is.
Other alternatives, such as storing a string "1y2m15d5h" (1 year, 2 months, 15 days and 5 hours. Extreme example) would render queries impossible, such as searching for all trials with duration longer than 1 year (there might even be some specified as 14m, which is also more than a year).
Possible solutions are:
Adding 4 columns just for this (years, months, days, hours). It would make queries much more complex than we wanted.
Forgetting support for durations in hours (probably a very uncommon case) and using "total number of days" instead. Unfortunately, this can't accurately represent a year, because of leap years (365 days might not be a full year).
Do you have other possible solutions for this? A readable, easily queryable format for a time/date offset? We couldn't find a concrete answer after searching the web.
As stated by myself and others in the comments, simplify the duration values so that you store it either in hours or days, ignore the longer months and leap years for the sake of simplicity.
This solution stores duration in days, to keep it simple for your queries for the UI and has a computed value for hours to be used in the calculation of the end date, although you may not need both columns, I'm just presenting them as an option:
CREATE TABLE #trials
(
Customer NVARCHAR(20),
StartTime DATETIME,
TrialDays INT,
TrialHours AS(TrialDays*24),
EndTime DATETIME
)
INSERT INTO #trials
(
Customer,
StartTime,
TrialDays,
EndTime
)
VALUES
( N'Bob', GETDATE(), 1, NULL), -- 1 day
( N'Dave', GETDATE(), 7, NULL), -- 7 days
( N'Jon', GETDATE(), 30, NULL), -- 30 days
( N'Tom', GETDATE(), 90, NULL), -- 90 days
( N'Jim', GETDATE(), 365, NULL) -- 365 days (approximation of a year)
SELECT * FROM #trials AS t
UPDATE #trials
SET EndTime = DATEADD(HOUR, TrialHours, StartTime)
SELECT * FROM #trials AS t
DROP TABLE #trials
Produces:
Customer StartTime TrialDays TrialHours EndTime
Bob 2017-09-11 10:50:30.940 1 24 2017-09-12 10:50:30.940
Dave 2017-09-11 10:50:30.940 7 168 2017-09-18 10:50:30.940
Jon 2017-09-11 10:50:30.940 30 720 2017-10-11 10:50:30.940
Tom 2017-09-11 10:50:30.940 90 2160 2017-12-10 10:50:30.940
Jim 2017-09-11 10:50:30.940 365 8760 2018-09-11 10:50:30.940
I am saving year, month , day and hrs in xml columns and calculating expire date by retrieving data from this xml column-
create table temp (
ID int , Duration xml , StartDate datetime
)
go
insert into temp
select 1 ,'<Time><Yr>1</Yr><Mo>2</Mo><Dy>3</Dy><Hr>4</Hr></Time>' , getdate()
union all select 2 ,'<Time><Yr>2</Yr><Mo>2</Mo><Dy>3</Dy><Hr>3</Hr></Time>', getdate()
union all select 3 , '<Time><Yr>1</Yr><Mo>7</Mo><Dy>6</Dy><Hr>9</Hr></Time>', getdate()
union all select 3 , '<Time><Yr>0</Yr><Mo>11</Mo><Dy>22</Dy><Hr>15</Hr></Time>' , getdate()
SELECT ID , StartDate ,
t1.Duration.value('(/Time/Yr)[1]','int')YR ,
t1.Duration.value('(/Time/Mo)[1]','int')MO,
t1.Duration.value('(/Time/Dy)[1]','int')DY ,
t1.Duration.value('(/Time/Hr)[1]','int')HR ,
dateadd( HOUR , t1.Duration.value('(/Time/Hr)[1]','int'), dateadd( day ,t1.Duration.value('(/Time/Dy)[1]','int') ,dateadd(MONTH ,t1.Duration.value('(/Time/Mo)[1]','int') , dateadd(year, t1.Duration.value('(/Time/Yr)[1]','int') , StartDate) ))) as EndTime
FROM temp t1

Business Week Group in SQL (instead of calendar week)

My requirement is that I want to find business-week-ending (not the calender week) given a DATE column from the sales table in MSSQL.
Using different techniques I was able to find the [Calender] week-endings (and week-starting) dates corresponding to DATE in the table.
Since our business week ends on Wednesday [DOW 3 or 4 depending when the week started], I tried to deduct number of days from the week ending dates to pull it back to Wednesday. The idea did work pretty good with a flaw. Works fine as long as the Date in the table is greater than DOW 3 or 4. Any suggestion?
SELECT DateAdd(wk, DateDiff(wk, 0, Recons_Sales_Details.Recons_Date), 0) + 2
You need to look into SET DATEFIRST to do this:
SET DATEFIRST 4 --4 is Thursday week start
SQL Fiddle Demo

TSQL Determine every other Friday from a "seed" date

Greetings StackOverflow Wizards.
SQL datetime calculations always give me trouble. I am trying to determine if an employee's hiredate fell between the last payday of that month and the first of the next month. (I.e. did they get a paycheck in their hire month.
Knowns:
I know our paydays are every other Friday.
I know 01/02/1970 was a Payday, and that date precedes the longest
active employee we have.
I know the hire date of each active employee (pulled from table).
I know (can calculate) the first of the month following the hire
date.
What I cannot seem to wrap my head around is how to use that seed date (01/02/1970) with datediff, dateadd, datepart, etc. to determine if there is a pay date between the hire date in question and the first of the following month.
In pseudo-code, here is what I'm trying to do:
declare #seedDate datetime = '01/02/1970' -- Friday - Payday seed date from which to calculate
declare #hireDate datetime = '09/26/2008' -- this date will actually be pulled from ServiceTotal table
declare #firstOfMonth datetime = DATEADD(MONTH, DATEDIFF(MONTH, 0, #hireDate) + 1, 0) -- the first of the month following #hireDate
declare #priorPayDate datetime -- calculate the friday payday immediately prior to #firstOfMonth
if #priorPayDate BETWEEN #hireDate AND #firstOfMonth
begin
-- do this
end
else
begin
-- do that
end
Using the hard-coded #hireDate above, and the #seedDate to determine every-other-Friday paydays, I know that there was a payday on 9/19/2008 and not another one until 10/03/2008, so the boolean above would be FALSE, and I will "do that" rather than "do this." How do I determine the value of #priorPayDate?
In all my databases where there is a lot going on with dates I create a table with colums for date,day, weekday,month,weeknr,dayof month, etc etc. I then use a procedural programming language or a bunch of handwritten sql to populate this table with every day for a large range of years say 1970 to 2200.
I pack this table 100% and index it heavily. You can then simply join any date to this table and do complex date stuff with simple where clause. So basically you pre calculate a helper table. maybe in you case you add a column to the date helper table with friday since seed column.
hope that makes sense.
Taking a DATEDIFF for days between your #seedDate and #firstOfMonth will give you a total number of days, which you can modulus by number of days between pay periods (14) to get number of days from the last pay period to the #firstOfMonth. You'll run into problems when the 1st is a payday (e.g. next month), which makes a CASE statement necessary:
DECLARE #priorPayDate DATETIME
SET #priorPayDate = CASE
WHEN DATEDIFF(dd, #seedDate, #firstOfMonth) % 14 = 0
THEN DATEADD(dd, -14, #firstOfMonth)
ELSE DATEADD(dd, -(DATEDIFF(dd, #seedDate, #firstOfMonth) % 14), #firstOfMonth)
END

Resources