Re: Does Codd's view of a relational database differ from that ofDate&Darwin?[M.Gittens]

Alexandr Savinov wrote:
>
> Update and delete-insert should be fundamentally different operations
> for the following reason. Update deals with object semantics, i.e., how
> they are characterized by other *existing* objects. In particular,
> update does not change the number of objects and their references.
> Delete-insert is a life-cycle management operation that has nothing to
> do what objects mean. Instead, it creates or deletes references, which
> reflect object existence. As I already mentioned somewhere in this forum
> we need to distinguish two layers:
> - object semantics (field values), and
> - object representation (references).
> If they are not separated like in the RM then there is no need to
> separate the operations of update and delete-insert (at fundamental
> level). Thus in the RM, I agree, these two operations are quite
> comparable. But in this case we have numerous problems.

There are advantages and disadvantages to distinguishing between value and identity, and there are advantages and disadvantages to discarding the concept of identity in favor of pure value semantics. It is not a simple "which is better" situation.

My perspective is that one significant disadvantage to the OO approach is the way that it puts little bits of state everywhere in your code, such that it is hard to manage or think about in any kind of central way. But I do not take the pure FP perspective that state must be completely eliminated; state is an important part of what we work with programs for.

I think the RM coupled with a heavily FP-influenced perspective is a sweet spot. State is limited to a few easily identified areas; code is (largely, depending on the model) stateless.

I don't think that the OO concept of object identity is worth the complexity and state-management difficulties it introduces. Others may think differently.

Marshall Received on Fri Jul 08 2005 - 19:32:18 CEST