sql server - SQL Recursive CTE: Finding objects linked by property

Question

I'm just trying to understand CTE and recursion to solve an issue that I would previously have used a cursor for.

create table ##ACC (
AccNo int,
Property char
)

Insert into ##ACC 
VALUES (1,'A'),(1,'B'),(2,'A'),(2,'C'),(3,'C'),(4,'D')

What I'm trying to achieve is to get a list of all AccNo's, and all AccNo's they're related to via Property. So my expected results are

PrimaryAccNo | LinkedAccNo
1 | 1
1 | 2
1 | 3
2 | 1
2 | 2
2 | 3
3 | 1
3 | 2
3 | 3
4 | 4

I've attempted the following code and variations but I either get 4 results (PrimaryAccNo=LinkedAccNo) only or I hit 100 recursions.

WITH Groups(PrimaryAccNo, LinkedAccNo)
AS
(
Select distinct AccNo, AccNo from ##ACC

UNION ALL
Select g.PrimaryAccNo, p.AccNo from
##ACC p inner join Groups g on p.AccNo=g.LinkedAccNo
inner join ##ACC pp on p.Property=pp.Property
where p.AccNo<> pp.AccNo
)
Select PrimaryAccNo,LinkedAccNo 
from Groups

What am I doing wrong?

Answer 1

You're running into an infinite loop caused by cycles within your data, e.g.: 1 > 2 > 3 > 2 > ... . The solution is to keep track of the rows that have already been "consumed". Due to limitations in CTEs, this has to be done by including the history within each CTE row, e.g. by assembling the path followed to arrive at each row. You can uncomment the , Path on the final select to see what is going on.

-- Sample data.
declare @ACC as Table ( AccNo Int, Property Char );
insert into @ACC values
  ( 1, 'A' ), ( 1, 'B' ), ( 2, 'A' ), ( 2, 'C' ), ( 3, 'C' ), ( 4, 'D' );
select * from @ACC;

-- Recursive CTE.
with Groups as (
  select distinct AccNo, AccNo as LinkedAccNo,
    Cast( '|' + Cast( AccNo as VarChar(10) ) + '|' as VarChar(1024) ) as Path
    from @ACC
  union all
  select G.AccNo, A.AccNo, Cast( Path + Cast( A.AccNo as VarChar(10) ) + '|' as VarChar(1024) )
    from Groups as G inner join -- Take the latest round of new rows ...
      @ACC as AP on AP.AccNo = G.LinkedAccNo inner join -- ... and get the   Property   for each ...
      @ACC as A on A.Property = AP.Property -- ... to find new linked rows.
      where G.Path not like '%|' + Cast( A.AccNo as VarChar(10) ) + '|%' )
  select AccNo, LinkedAccNo -- , Path
    from Groups
    order by AccNo, LinkedAccNo;