>
>Here I'd like to see perhaps a comparison chart or listings of the pros and cons of each approach. We hear "data tier" vs. "business logic tier", but seldom see *performance* comparisons of the different approaches.
>
>In Erik's example, is there a performance hit if he puts the table opening and access code into a separate data tier component? How does performance weigh against purity of design and ease of swapping back-ends? Again, this is the kind of info that I'd like to see quantitatively compared in your book -- real-life balanced against theory.

I'll throw in my $.02. In the Mere Mortals framework, the business object and data access object are separated, but are grouped together in a container. I can't say I ever noticed any appreciable difference in speed between that and a direct approach where my bizobj opened the tables. I have a tendency to try a bunch of different techniques in my programs (which is going to make it interesting to whoever inherits my code ;-) ) to get a feel for them. Since I'm not a full-time programmer I try as many things as I can when I have the chance. I've tried both ways and separating the objects definitely "feels" right. It's more work initially but it's a more elegant design - I think it's worth whatever minor overhead hit you might take in instancing it.

If you haven't looked at it already, read up on the hook pattern over on Steven Black's site.