Greetings - I just cooked up my first example objects with the current .5x stream of SQLObject and was nicely rewarded. I love the table schema generation - one of the things I've always loathed was maintaining the table/object mapping metadata in two different places.
I'm really shocked at how liberating this is! ;)
I've done a number low to middle complexity solutions using "ORM" (tuxweb.de) as well as Titus Brown's Cucumber solution (which makes elegant use of Postgresql inheritance, but is therefore not portable to other systems. I rather like the API in SQLObject; a quick comparison of current performance shows it to be somewhat more efficient at some common tasks, and easy to side step when required.
Just going to check out the new work in subversion and see how that compares.