What I would find way more useful would be an abstraction that removes the SQL language itself, i.e. a relational database library. That would have some advantages:

* Better coherence with C++ data types. * Possibility to support non-SQL relational databases. * Make it easier to construct object/relational translation abstractions.

Yes, absolutely. I would love to see a non-SQL C++-interfaced relational database library. With my query-is-a-functor class the SQL query string can be hidden away, and at the point of use you can fantasize that it's not doing SQL.... But in order to be useful, such a library would need to offer all of the important features of current databases, i.e. good performance with huge on-disk data sets, transactions, client/server operation, concurrency, and so on. To some extent these requirements can be usefully addressed as individual problems. So if someone is looking for something to do, how about starting with containers, iterators and allocators for huge on-disk data structures. This could be based on the existing shared memory containers, but using mmap()ed windows into files. Then we'd probably want some sort of dynamic algorithm selection ("query planning") mechanism. Regards, Phil.