Re: What is a Relationship !?

Tom Hester wrote:

>>By that definition most of the relations in a database could be
>>classified as relationships, I guess...
>>
>>
>No, first of all; there is the question of levels of abstraction. I suppose
>that you mean that the relations in a database model or describe a relation?
>But in fact, some will describe entities and some relationships or they may
>in fact not describe either or describe only parts of either. It depends on
>how the modeling was done, among other things.
>
>
>>And, further, a relationship could become an entity just by giving it a
>>unique identifier.
>>
>>
>No, it also must exist in the domain of discourse. Admittedly a squishy
>concept but one that is common in modeling--both data modeling and logical
>modeling.
>
>
>>job(person_id, company_id, etc...) //current job(s) of person
>>
>>Now would 'job' be an entity or a relationship?
>>
>>
>Job would be a relationship.
>
>
>>What if I add start_date?
>>
>>
>As I said, relationships can have attributes.
>
>
>>job(person_id, company_id, start_date, etc...) //job history of person
>>
Yes, but since this saves the history start_date is part of the key (you can go back to
the same company at a later time; sorry, forgot to mention that) Still a relationship?

>>What if I add a surrogate key?
>>
>>
>I would say that you are doing relational modeling and not ER modeling.
>

OK, but perhaps the database was for a labour pension company, whose main function is to track
peoples employment history over the years and the concept of an employment is essential. Is it a relationship then? (OK: this must be the "squishy concept" you mentioned earler)

My point here is that I fail to understand the added value of EM diagrams (as opposed to pure
entity diagrams). And if the distinction is basically arbitrary (in some, if not most cases) then why
waste energy trying to decide which one it is?

If we can get along with one concept (entity) instead of two (entity, relationship), why should we introduce this extra concept??

The danger I see in E/R modelling is that N:M relationships are "left alone" for too long in the process
resulting in nasty surprices down the road.

best regards,
Lauri Pietarinen Received on Thu Nov 20 2003 - 23:36:45 CET