> In my experience with various ORMs and object persistence
> infrastructure, the challenge is always how to decompose a non-trivial
> data structure either into a flat representation suitable for key-value
> datastores, or into inter-related tables suitable for relational
> datastores. Although tools such as SQLAlchemy provide convenience
> functions for simple structures, inevitably there is substantial design
> effort involved in representing hierarchical data types, e.g. record of
> records or lists. So I feel that the problem of hierarchical data
> decomposition is much more fundamental than abstracting the operational
> details of the storage backend.

+1. One important subproject of Schemepersist is to design a
SQLAlchemy-like representation, which we started last year.

To be honest, relational databases are a pain in the ass to work with
IMHO. I've never come across any problem (trivial or non-trivial) for
which a relational representation is the natural one. Simple relational
queries are pretty nice, but once you get to joins it's baffling and
just gets worse from there.

That being said, I fully agree with John that SQLite is a very nice
storage backend (as are other RDBMSes) and should be supported.