Intersect Select Statements on Specific Columns - sql

Question

I've a table of SalesDetails, looking like this:
InvoiceID, LineID, Product
1,1,Apple
1,2,Banana
2,1,Apple
2,2,Mango
3,1,Apple
3,2,Banana
3,3,Mango
My requirement is to return rows where an Invoice contained sales of both: Apple AND Banana, but if there are other products on such an invoice, I don't want those.
So the result should be:
1,1,Apple
1,2,Banana
3,1,Apple
3,2,Banana
I tried the following:
Select * from SalesDetails where Product = 'Apple'
Intersect
Select * from SalesDetails where Product = 'Banana'
Didn't work, because it seems Intersect needs to match all the columns.
What I'm hoping to do is:
Select * from SalesDetails where Product = 'Apple'
Intersect ----On InvoiceID-----
Select * from SalesDetails where Product = 'Banana'
Is there a way to do this?
Or do I have to first Intersect on InvoiceIDs only using my criteria, then select the rows of those InvoiceIDs where the criteria is matched again, I.e.:
Select * From SalesDetails
Where Product In ('Apple', 'Banana') And InvoiceID In
(
Select InvoiceID from SalesDetails where Product = 'Apple'
Intersect
Select InvoiceID from SalesDetails where Product = 'Banana'
)
Which seems somewhat wasteful as it's examining the criteria twice.

Answer 1

Okay this time I've managed to get reuse of the Apple/Banana info by using a CTE.
with sd as (
Select * from SalesDetails
where (Product in ('Apple', 'Banana'))
)
Select * from sd where invoiceid in (Select invoiceid from
sd group by invoiceid having Count(distinct product) = 2)
SQL Fiddle

Answer 2

Do it with conditional aggregation:
select *
from SalesDetails
where product in ('apple', 'banana') and invoiceid in(
select invoiceid
from SalesDetails
group by invoiceid
having sum(case when product in('apple', 'banana') then 1 else 0 end) >= 2)

Answer 3

I think OP's suggestion is about the best one can do. The following might be faster, although I expect the difference to be slight and I have not done any benchmarking.
Select * From SalesDetails
Where Product ='Apple' And InvoiceID In
(
Select InvoiceID from SalesDetails where Product = 'Banana'
)
union all
select * from SalesDetails
Where Product ='Banana' And InvoiceID In
(
Select InvoiceID from SalesDetails where Product = 'Apple'
)

Answer 4

A self-join will solve the problem.
SELECT T1.*
FROM SalesDetails T1
INNER JOIN SalesDetails T2 ON T1.InvoiceId = T2.InvoiceId
AND (T1.Product = 'Apple' AND T2.Product = 'Banana'
OR T1.Product = 'Banana' AND t2.Product = 'Apple')

Answer 5

declare #t table (Id int,val int,name varchar(10))
insert into #t (id,val,name)values
(1,1,'Apple'),
(1,2,'Banana'),
(2,1,'Apple'),
(2,2,'Mango'),
(3,1,'Apple'),
(3,2,'Banana'),
(3,3,'Mango')
;with cte as (
select ID,val,name,ROW_NUMBER()OVER (PARTITION BY id ORDER BY val)RN from #t)
,cte2 AS(
select TOP 1 c.Id,c.val,c.name,C.RN from cte c
WHERE RN = 1
UNION ALL
select c.Id,c.val,c.name,C.RN from cte c
WHERE c.Id <> c.val)
select Id,val,name from (
select Id,val,name,COUNT(RN)OVER (PARTITION BY Id )R from cte2 )R
WHERE R = 2

Answer 6

Other was is to do PIVOT like this:
DECLARE #DataSource TABLE
(
[InvoiceID] TINYINT
,[LineID] TINYINT
,[Product] VARCHAR(12)
);
INSERT INTO #DataSource ([InvoiceID], [LineID], [Product])
VALUES (1,1,'Apple')
,(1,2,'Banana')
,(2,1,'Apple')
,(2,2,'Mango')
,(3,1,'Apple')
,(3,2,'Banana')
,(3,3,'Mango');
SELECT *
FROM #DataSource
PIVOT
(
MAX([LineID]) FOR [Product] IN ([Apple], [Banana])
) PVT
WHERE [Apple] IS NOT NULL
AND [Banana] IS NOT NULL;
It will give you the results in this format, but you are able to UNVPIVOT them if you want:
Or you can use window function like this:
;WITH DataSource AS
(
SELECT *
,SUM(1) OVER (PARTITION BY [InvoiceID]) AS [Match]
FROM #DataSource
WHERE [Product] = 'Apple' OR [Product] = 'Banana'
)
SELECT *
FROM DataSource
WHERE [Match] =2

Answer 7

First, you want to COUNT the number of rows per InvoiceID that matched the criteria Product = 'Apple' or 'Banana'. Then do a SELF-JOIN and filter the rows such that the COUNT must be >= 2, or the number of Products in your critera.
SQL Fiddle
SELECT sd.*
FROM (
SELECT InvoiceID, CC = COUNT(*)
FROM SalesDetails
WHERE Product IN('Apple', 'Banana')
GROUP BY InvoiceID
)t
INNER JOIN SalesDetails sd
ON sd.InvoiceID = t.InvoiceID
WHERE
t.CC >= 2
AND sd.Product IN('Apple', 'Banana')

Answer 8

WITH cte
AS
(
SELECT *
FROM [dbo].[SalesDetails]
WHERE [Product]='banana')
,cte1
AS
(SELECT *
FROM [dbo].[SalesDetails]
WHERE [Product]='apple')
SELECT *
FROM cte c INNER JOIN cte1 c1
ON c.[InvoiceID]=c1.[InvoiceID]

Answer 9

Here is a method using window functions:
select sd.*
from (select sd.*,
max(case when product = 'Apple' then 1 else 0 end) over (partition by invoiceid) as HasApple,
max(case when product = 'Banana' then 1 else 0 end) over (partition by invoiceid) as HasBanana
from salesdetails sd
) sd
where (product = 'Apple' and HasBanana > 0) or
(product = 'Banana' and HasApple > 0);

Answer 10

If you only want to write the condition once and are sure that each Product will only be once in any Order, you can use this:
SELECT * FROM (
SELECT InvoiceID, Product
,COUNT(*) OVER (PARTITION BY InvoiceID) matchcount
FROM SalesDetails
WHERE Product IN ('Apple','Banana') ) WHERE matchcount = 2;

Answer 11

This is what I ended up using, inspired by #Leon Bambrick:
(Expanded a little to support multiple products in the criteria)
WITH cteUnionBase AS
(SELECT * FROM SalesDetails
WHERE Product IN ('Apple Red','Apple Yellow','Apple Green','Banana Small','Banana Large')),
cteBanana AS
(SELECT * FROM cteUnionBase
WHERE Product IN ('Banana Small','Banana Large')),
cteApple AS
(SELECT * FROM cteUnionBase
WHERE Product IN ('Apple Red','Apple Yellow','Apple Green')),
cteIntersect AS
(
SELECT InvoiceID FROM cteApple
Intersect
SELECT InvoiceID FROM cteBanana
)
SELECT cteUnionBase.*
FROM cteUnionBase INNER JOIN cteIntersect
on cteUnionBase.InvoiceID = cteIntersect.InvoiceID