On Nov 12, 2003, at 8:05 PM, Mark A. Schwenk wrote:

> I'd like to look at any Tantalus documentation and examples that you
> could resurrect. I still think the code has a number of things going 
> for
> it.

Ok, I've created a new home page for Tantalus and updated the SqueakMap 
entry to point to it. The documentation, such as it is, is there as 
well.

http://www.wiresong.ca/tantalus/

> What sorts of problems did you run into when using Tantalus?

Well, it's been quite a while, but here's what I remember.

One problem is that to-many relations are difficult to handle, 
especially when it comes to caching. From the Smalltalk end, they 
should be modelled as collections, but in the database the references 
point the wrong way. It's real easy for the inverse to-one relations to 
get changed, and leave the collection out of date.

Another was that generating SQL was pretty ugly, particularly when 
dealing with database-specific bits. We ended up with a fairly 
reasonable class hierarchy for that, but it wasn't very satisfying, and 
I wouldn't enjoy adding support for a new database.

Finally setting up the mappings programmatically is a real hassle. We 
did some work on building mappings from EOF files, but I left 
Whistler.com before that was released.

Roe addresses the first two issues nicely, I think, by providing an 
elegant model for SQL queries. If a mapping framework were built to 
take advantage of it, we might be able to explore the idea of doing 
away with the caching altogether, which would solve the first problem.

Colin