Re: Sensible and NonsenSQL Aspects of the NoSQL Hoopla

On 2013-08-28, karl.scheurer_at_o2online.de <karl.scheurer_at_o2online.de> wrote:
> Am Mittwoch, 28. August 2013 05:00:46 UTC+2 schrieb James K. Lowden:
>> On Mon, 26 Aug 2013 03:20:44 -0700 (PDT)
>> karl.scheurer_at_o2online.de wrote:
>> Many programming languages had data structures when the relational
>> model was invented. Not all, of course, not Forth for example. But
>> Cobol, Fortran, Algol, and PL/1 did.
>>
> Sorry, data structures like records and arrays are data types in our
> terminology. We call linked lists, tree, graphs, stacks, queues ...
> (dynamic) data structures. These structures are not part of standard
> libraries before C++. The next step after data structues are objects.
> (data structures with special behaviours). We started objects with
> homegrown matrix- and complex- number objects for engineering calculations.
> Relations can be implemented as objects as I know at least one system
> complete done with objects.
>
> I challenge a non object implementation for relations with satisfying
> Codd' constraints (row-, column order are insignificant).

That's because you can't get hold of the idea that physical storage is irrelevant (as long as it works and is efficient enough). What the constraints mean in practice is that when you retrieve a set of rows you can make _no_ assumptions about the order they will be in unless you explicitly specify an order. Column order is irrelevant because you have to specify which columns you want and then (usually) get them in the order specified. If you use SELECT * you are back to the order being unpredictable (though in practice usually predictable in any single RDBMS, but this is an artifact of the implementation of the RDBMS, not a requirement of relational theory).

Eric

-- 
ms fnd in a lbry