Re: The Theoretical Foundations of the Relational Model

On 25 Jun 2002 00:41:13 +0200, hidders_at_hcoss.uia.ac.be (Jan.Hidders) wrote:

>> (sorting) has zero effect on logical operations, it has positive benefit to
>>whole applications, so why not allow it?
>
>It is not true that it has zero effect on logical operations. If the
>database has to maintain information about the order your updates become
>more expensive.

That is not a logical issue.

And as far as efficiency, if the sort is going to be done anyway, say as part of a view, then why not make the specification static?

>It is also not true that things always become easier for the user. He or she
>now has to decide which tables are ordered and which are not.

It probably would not be advisable to do it often. However, most SQLServer designers consider it an important art form to use the "clustered index", which is in effect a physical ordering of the table. Myself, I never, ever use it, consider it a totally obsolete remnant feature from days before RAID and gigabyte RAM processors.

>If the
>ordering is important then you also need to introduce special update
>operations such as insert-at(position) and move(position, position). You
>will have to introduce special functions and predicates such as comes-before
>and is-at(position) that operate on this order. Do you really believe that
>this is going to make things simpeler for the user?

None of that is necessary at the user level, and it is anyway already provided at the b-tree management level internal to the database.

Come on, you know that. There just seems to be something about the very suggestion of adding order as a property of views and tables, that causes reflex reactions of fear and loathing in otherwise sane and sober relational database experts.

Joshua Stern Received on Tue Jun 25 2002 - 01:28:32 CEST