Carbonado, another approach to Java persistence, by Amazon

This looks like quite a tight marriage of objects and relations. Joins are specified in the definitions of the joined classes, in much the same way that foreign keys might be specified in an SQL schema, but are apparently automatically used when querying. Carbonado provides its own query implementation, so it does not even require a SQL backend (although integration concerns may mandate the use of one).

I have dreamed about an approach like this for a while. Now I can dream instead about making the time to study it...