SQL cartesian join problem

Question

I have three tables

• A: A.pID primary key, A.Name nvarchar(250)
• B: B.pID primary key, B.Name nvarchar(250)
• C: C.pID primary key, C.Name nvarchar(250)

There is a m to n relation between A and B (table lA_B with primary key lA_B.pID and .pInstanceA Foreign key to table A and .pInstanceB Foreign key to table B)

There is a m to n relation between A and C (table lA_C with primary key lA_C.pID and .pInstanceA Foreign key to table A and .pInstanceB Foreign key to table C)

• A1 is in relation with B1, B2 and C1
• A2 is in relation with B3 and C2, C3
• A3 is in relation with B4
• A4 is in relation with C4
• A5 has no relation

Here is my SQL:

CREATE TABLE [dbo].[A]( [pID] [bigint] NOT NULL, [Name] [nvarchar](250) NULL )
CREATE TABLE [dbo].[B]( [pID] [bigint] NOT NULL, [Name] [nvarchar](250) NULL )
CREATE TABLE [dbo].[C]( [pID] [bigint] NOT NULL, [Name] [nvarchar](250) NULL)
CREATE TABLE [dbo].[lA_B]( [pID] [bigint] NOT NULL, [pInstanceA] [bigint] NULL, [pInstanceB] [bigint] NULL )
CREATE TABLE [dbo].[lA_C]( [pID] [bigint] NOT NULL, [pInstanceA] [bigint] NULL, [pInstanceB] [bigint] NULL )

INSERT INTO [dbo].[A] ([pID] ,[Name]) VALUES (1,'A1')
INSERT INTO [dbo].[A] ([pID] ,[Name]) VALUES (2,'A2')
INSERT INTO [dbo].[A] ([pID] ,[Name]) VALUES (3,'A3')
INSERT INTO [dbo].[A] ([pID] ,[Name]) VALUES (4,'A4')
INSERT INTO [dbo].[A] ([pID] ,[Name]) VALUES (5,'A5')

INSERT INTO [dbo].[B] ([pID] ,[Name]) VALUES (1,'B1')
INSERT INTO [dbo].[B] ([pID] ,[Name]) VALUES (2,'B2')
INSERT INTO [dbo].[B] ([pID] ,[Name]) VALUES (3,'B3')
INSERT INTO [dbo].[B] ([pID] ,[Name]) VALUES (4,'B4')

INSERT INTO [dbo].[C] ([pID] ,[Name]) VALUES (1,'C1')
INSERT INTO [dbo].[C] ([pID] ,[Name]) VALUES (2,'C2')
INSERT INTO [dbo].[C] ([pID] ,[Name]) VALUES (3,'C3')
INSERT INTO [dbo].[C] ([pID] ,[Name]) VALUES (4,'C4')

INSERT INTO [dbo].[lA_B] ([pID],[pInstanceA],[pInstanceB])   VALUES   (1,1,1)
INSERT INTO [dbo].[lA_B] ([pID],[pInstanceA],[pInstanceB])   VALUES   (2,1,2)
INSERT INTO [dbo].[lA_B] ([pID],[pInstanceA],[pInstanceB])   VALUES   (3,2,3)
INSERT INTO [dbo].[lA_B] ([pID],[pInstanceA],[pInstanceB])   VALUES   (4,3,4)

INSERT INTO [dbo].[lA_C] ([pID],[pInstanceA],[pInstanceB])   VALUES   (1,1,1)
INSERT INTO [dbo].[lA_C] ([pID],[pInstanceA],[pInstanceB])   VALUES   (2,2,2)
INSERT INTO [dbo].[lA_C] ([pID],[pInstanceA],[pInstanceB])   VALUES   (3,2,3)
INSERT INTO [dbo].[lA_C] ([pID],[pInstanceA],[pInstanceB])   VALUES   (4,4,4)

this query:

SELECT
  A.Name AS A, 
  B.Name AS B, 
  C.Name AS C
FROM
  A  
  left JOIN lA_B  ON (A.pID = lA_B.pInstanceA)
  left JOIN B     ON (B.pID = lA_B.pInstanceB)
  left JOIN lA_C  ON (A.pID = lA_C.pInstanceA)
  left JOIN C     ON (C.pID = lA_C.pInstanceB) 

returns

A1      B1      C1
A1      B2      C1
A2      B3      C2
A2      B3      C3
A3      B4      NULL
A4      NULL    C4
A5      NULL    NULL 

And now the question :-) how to query to receive

A1      B1      NULL
A1      B2      NULL
A1      NULL    C1
A2      B3      NULL
A2      NULL    C2
A2      NULL    C3
A3      B4      NULL
A4      NULL    C4
A5      NULL    NULL 

The problem is that when I make the join both with B and with C the result has all the combinations of B C. How can I eliminate this?

Answer 1

You might be able to do that with a UNION:

SELECT A.Name AS A, B.Name AS B, NULL AS C 
FROM A
left JOIN lA_B ON (A.pID=lA_B.pInstanceA) 
left JOIN B ON (lA_B.pInstanceB=B.pID) 
UNION
SELECT A.Name AS A, NULL AS B, C.Name AS C 
FROM A
left JOIN lA_C ON (A.pID=lA_C.pInstanceA) 
left JOIN C ON (lA_C.pInstanceB=C.pID)

The first part selects all combinations of A and B, the second part all combinations of A and C.

If you wish to filter out rows like (A4,NULL,NULL) because there is already a row (A4,NULL,C4), try this query:

SELECT A.Name AS A, B.Name AS B, NULL AS C 
FROM A
LEFT JOIN lA_B ON (A.pID=lA_B.pInstanceA) 
LEFT JOIN B ON (lA_B.pInstanceB=B.pID) 
WHERE b.name is not null 
    or not exists(select * from lA_C where A.pID=lA_C.pInstanceA) 
UNION
SELECT A.Name AS A, NULL AS B, C.Name AS C 
FROM A
LEFT JOIN lA_C ON (A.pID=lA_C.pInstanceA) 
LEFT JOIN C ON (lA_C.pInstanceB=C.pID)
WHERE c.name is not null 
ORDER BY A,B,C

For the join on B, this says to include rows that have a match in B, or for which there is no match in C. The join on C includes only rows which match in C. Rows that do not match either would get included from the join on B.

Note that UNION filters out duplicate rows, like DISTINCT. To include every row, you can use UNION ALL.

Answer 2

I think this does it:

select a,
       case when b='zzz' then null else b end as b,
       case when c='zzz' then null else c end as c
from (SELECT  A.Name AS A
             ,b.Name as b
             ,'zzz' as c
      FROM  A    
      JOIN lA_B  ON (A.pID = lA_B.pInstanceA)  
      JOIN B     ON (B.pID = lA_B.pInstanceB)  
      union
      select  a.Name
             ,'zzz'
            ,c.NAme
      from A
      left JOIN lA_C  ON (A.pID = lA_C.pInstanceA)  
      left JOIN C     ON (C.pID = lA_C.pInstanceB)) as a