Select aggregate ignores where cause - sql

Question

I'm trying to transform an existing view into a format I can work with.
The view vw_temp_appHoursLastTwoEntries looks like this:
RowNumber | PersNr | Client | Localtion | Agent | Date | Calweek | Year
----------+--------+--------+-----------+-------+------------+---------+------
1 | 123 | 1 | 1 | ag-01 | 2020-01-01 | 1 | 2021
2 | 123 | 1 | 1 | ag-01 | 2020-01-03 | 1 | 2021
1 | 9999 | 1 | 4 | ag-01 | 2020-01-01 | 1 | 2021
2 | 9999 | 1 | 4 | ag-01 | 2020-01-07 | 1 | 2021
I need this data in a different format that would look like this:
PersNr | Client | Localtion | Agent | minDate | MaxDate | DateDiff | Calweek | Year
-------+--------+-----------+-------+------------+------------+----------+---------+-------
123 | 1 | 1 | ag-01 | 2020-01-01 | 2020-01-03 | 3 | 1 | 2021
9999 | 1 | 4 | ag-01 | 2020-01-01 | 2020-01-07 | 7 | 1 | 2021
in the original format, one person has only two rows (RowNumber 1 and 2). I'd like to match each column and have the min and max date as well as the difference in a new view.
my Code:
select a.persnr, a.client, a.location, a.agent, a.calweek, a.year,
max(a.date) as maxdate, min(b.date) as mindate
, DATEDIFF(day,a.date,b.date) as dDiff
from vw_temp_appHoursLastTwoEntries a
left join vw_temp_appHoursLastTwoEntries b on
a.persnr = b.persnr and a.client = b.client and
a.agent = b.agent and a.date = b.date
where a.date != b.date and DATEDIFF(day,a.date,b.date) != 0
or (a.date is not null and b.date is not null)
group by a.persnr, a.client, a.location, a.agent, a.calweek, a.year, DATEDIFF(day,a.date,b.date)
The issue:
I'm currently getting back values where it seems like the where cause does not take effect but I don't understand why.
a.date != b.date should not return rows where min- and maxdates are the same. The datediff does not return any other value then 0 even when the min- and maxdates are different.

Answer 1

Pretty sure this is what you want:
declare #Test table (RowNumber int, PersNr int, Client int, Localtion int, Agent varchar(5), [Date] date, Calweek int, [Year] int);
insert into #Test (RowNumber, PersNr, Client, Localtion, Agent, [Date], Calweek, [Year])
values
(1, 123, 1, 1, 'ag-01', '2020-01-01', 1, 2021),
(2, 123, 1, 1, 'ag-01', '2020-01-03', 1, 2021),
(1, 9999, 1, 4, 'ag-01', '2020-01-01', 1, 2021),
(2, 9999, 1, 4, 'ag-01', '2020-01-07', 1, 2021);
select a.PersNr, a.Client, a.Localtion, a.Agent, a.Calweek, a.[Year]
, max(a.[date]) as maxdate
, min(b.[date]) as mindate
, abs(datediff(day,a.[date],b.[date])) as dDiff
from #Test a
left join #Test b on
a.persnr = b.persnr and a.client = b.client and
a.agent = b.agent --and a.[date] = b.[date]
where (/*a.[date] != b.[date] and*/ datediff(day,a.[date],b.[date]) != 0)
and /* not OR */ (a.[date] is not null and b.[date] is not null)
group by a.persnr, a.client, a.Localtion, a.agent, a.calweek, a.[Year], abs(datediff(day,a.[date],b.[date]));
Returns:
PersNr
Client
Localtion
Agent
Calweek
Year
maxdate
mindate
dDiff
123
1
1
ag-01
1
2021
2020-01-03
2020-01-01
2
9999
1
4
ag-01
1
2021
2020-01-07
2020-01-01
6
As Giorgos points out, you don't want to join on a.[date] = b.[date] because your where clause specifically filters that condition out.
The main issue was using OR instead of AND, you want to ensure that both date values are not null so that is an AND condition.
I am also assuming that dDiff is for debugging purposes only, which as you have it kept the rows from grouping, but you can group them by using the absolute value (abs).
You also don't need to test a.[date] != b.[date] because that is already true by virtue of datediff(day,a.[date],b.[date]) != 0.
Please use this form of DDL+DML (or a temp table) in future to provide sample data for us to work with (it gives you a minimal reproducible example also which is never a bad thing, because I picked up a number of typos in your query while copying it).