Posted by --CELKO-- on 11/03/53 11:43

This problem is harder than people think to do in the general case.
Ever have a course in Graph Theory? Remember some of the NP-Complete
problems at the end of that course?

Consider a set of paths between two nodes { A -> B, B -> A }. Pretty
easy to see that you can have an endless loop with just two tables. But
I will need to dtect ALL cycles of ANY size in any schema to avoid this
for the genral case

Consider a set of paths on one node { A -> A } , the smallest cycle
possible. You wipe out a whole table, set everything to one value or
hang in a loop. A.x changes A.y, and A.y changes A.x .. what does the
table look like after the constraints fire?

Consider a set of paths among three nodes { A -> B, A -> C, B-> C }
since C can be changed by both A and B, which one takes effect in a
declarative language whose statements are supposed to be independent of
an order of execution?

An early version of DB2 would allow this particular set of REFERENCES,
but the results were unpredictable -- the last change would persist.
The reason that you can do some of this with triggers is that they are
procedural and have a fixed order of execution. Of course the
optimizer cannot use them in a plan and it is non-declarative.

[Back to original message]