Re: Real world issue: Steps towards a unified method

>>Marshall ha scritto:

 Let's use this notation

   F1, F2, ... any table such that at least 1 field is involved in what

               you call a "nonidempotent" function, i.e., a function that

               is affected by possible value duplication (count, sum, avg,...)

   A1, A2, ... any other table

(Intuitively: F for forbid duplication, A for allow duplication). (I am still clear if I should also use a symbol for tables with no fields involved
in the report, but now I think that is probably useless.)

Let also use the notation used by Alexander for 1-N relationsip

   X

    \     means   X  -< Y   (one to many relationship)
     Y

See a couple of output of a possible automated method.

EXAMPLE 1. Pizza

Consider for instance

      A1    A2     A3
        \  /    \    /
         F1    F2

Output:
in this case the query will be:

   A1 A2 A2 A3
    \ / union \ /
     F1 F2

EXAMPLE 2. subtrans (pathological)

    A1

     \
      F1
       \
        F2

Output:
in this case there is no union that can make it "safely" there are 2 distinct "partitions" P1, P2 :

P1: P2: F2

    A1

     \
      F1

with a "cutting" relationship: F1
                                          \
                                         F2

possible queries in addition to

    A1

     \
      F1
       \
        F2

which might yeld wrong results, would only be:

    A1

     \
      F1

    F2

The presence of the "partition" should rise a possible alert to the user [this is the same as your "what about not to do it"]

The user should analyze the situation and decide what to do

By "systematic" method I mean a technique to give this kind of outputs *no matter how complicate is the pattern of tables and relationships *

I have some intuitive idea on how to do that..., but I would like to discuss it ...

-P Received on Sun Sep 17 2006 - 12:44:33 CEST