From: Lennon Day-Reynolds <rcoder@...>
Date: 2004-07-29T14:57:41+09:00
Subject: Re: Stupid ODBC!

On Thu, 29 Jul 2004 10:02:25 +0900, Nathaniel Talbott
<nathaniel@talbott.ws> wrote:
[...]
> Then I don't understand something about your request; you said you were
> looking for:
> > standard way of getting an identifier for the last row inserted
> If a row is expected to have an identifier, then it's going to have to
> have an identity column. I believe other databases, such as Postgres
> and MySQL, don't force you to have an identity column, either, so I
> don't see what makes SQL Server any more onerous than other databases
> (in this regard; I have lots of other reasons to hate it if you need
> some).
> 
> Am I missing some context here on Active Record?

The SQL Server concept of IDENTITY is different from a ROWID (as in
Oracle, Postgres, etc.), as it is an *optional* type, much like a
primary key is optional.

This is only an issue becuase of the requirements of Active Record;
namely, the fact that an Active Record object needs a persistent,
unique identifier, even if the underlying database table row doesn't
have one.

In Postgres et. al., that's easy, since the internal value that the
database engine uses to refer to a row is accessible via the
afore-mentioned DBMS-specific functionality. SQL Server still *uses* a
globally unique id to refer to each row in a database, it just doesn't
expose that value to client applications through ODBC.

Lennon