I was reading a thread on the cgiapp mailing list today from several of the core developers about developing a book on CGI::Application. In it, several mentioned that it might/should center around CGI::App and a handful of oft-used modules. One of those modules is Class::DBI.

I took a gander at Class::DBI over at CPAN, and it looks absolutely amazing, and at the same time perhaps too abstract. Basically, you create a number of packages and/or packages, one for each table you'll be using in your application, and one to establish your basic connection. Then, each package creates an object instance of the connection, and defines a number of properties: the name of the table, the columns you'll be using, and then the relations it has to other tables ( has_a( col_name => 'Package::Name'); has_many( col_name => 'Package::Name'); might_have(col_name => 'Package::Name'); ) etc.

Then you use the module/packages you need in your script, and you can then use object-oriented notation to do things like insert rows, update rows, search a table, select rows, etc. And it looks fairly natural.

I like the idea of data abstraction like this. I see a couple issues, however:

1. I don't like the idea of one package per table; that gets so abstract as to make development come to a stand-still, especially during initial development. However, once development is sufficiently advanced, I could see doing this, particularly for large projects; it could vastly simplify many regular DBI calls.
2. I like using SQL. If I need to debug why something isn't working when I interact with the database, I want to have absolute control over the language. Abstracting the SQL means I don't have that fine-grained control that helps me debug.

So, for now, I'll stick with straight DBI…. but this is an interesting avenue to explore.