Re: A searchable datastructure for represeting attributes?

--CELKO-- wrote:
> The short answer is "No" and the long answer is that you are
> committing what I call "The OO Programmer's Mistakes" now that I have
> had to clean it up a few times. Your first approach is a
> META-DATABASE model trying to wedge itself into a DATABASE. It just
> don't work!!

Reading this very nice thread, I can only support Joe's opinion.

A relational database is optimized for querying for relations. Storing all "objects" or "object attributes" into one table is always bound to become a desaster. Don't even think about it.

There is a very nice analogy on the OO programmers side: People modelling META-CLASSES in programming languages:

class MetaClass{
  String name;
  MetaAttribute[] attributes;
}

class MetaAttribute{
  String name;
  Object value;
}

In November I had to spend about 50 emails fighting a team of 4 developers to prevent the above.

The simple solution for both META-BLUNDERS: - Create your relational tables dynamically from your object attributes. Every database engine supports a CREATE TABLE statement. It's a little tricky to reengineer a table by copying the data to a temporary table, dropping the table and recreating it but that's only a days work. With this approach, some constraints will be tricky but your approach has none at all.
- Create your classes dynamically. Speaking for Java, you can either write .Java files and compile them or use a bytecode engineering library like BCEL.

> You want a persistent object store of some sort. I don't know if
> anyone hs such a thing that can be accessed easily, to be honest. Do
> not do this in SQL.

Here gos:

db4o - database for objects
http://www.db4o.com

Kind regards,
Carl Received on Fri Mar 01 2002 - 00:30:58 CET