Re: Conflicts in Relationships

bill_dev wrote:

> Jan, it appears something must be missing in our algorithm. I had no
> problem implementing it and I put a print statement in place of the
> check of the join condition. That way I can see that I'm walking the
> nodes correctly.
>

>>You said......all the join conditions have to be full on both
>>sides, i.e., they have to be *full* outer joins. All other join
>>conditions only have to full on one side.

What you call an outer join is what I call a full outer join or a join that is both full on both sides. What you call a left join is what I call a left outer join or a join that is only full on the left side, et cetera. Since outer joins are not really part of the relational model the terminology tends to vary a little.

> For example, in my 3 table query above join conditions like these
> denote a valid query:
>
> A Left Join C
> C Inner Join B
> A Left Join B
>
> In my code I load the 3 tables and 6 edges (3 edges in both directions)
> and I walk the graph with this output:

I would have used 1 undirected edge per join condition, but this will also work.

> (TableA,TableC) Left Outer Join
> *** (TableC,TableB) Inner Join
> (TableA,TableB) Left Outer Join
>
> At the asterisks the code would have stopped to report a conflict. Am I
> doing somethng wrong or do we need somethin more complex to handle the 4
> possible states of a join? Any ideas?

What is the problem? The join between B and C should indeed be a (full) outer join as is reported by the algorithm if you don't let it stop at the first conflict it sees. This is because it will report that the join should be full on the side of B and full on the side of C, so therefore a full outer join.

• Jan Hidders