I'd like to finish C subprocess implementations of SQLite and Postgres.
Harold is interested in doing FFI and/or socket versions.

Can someone start throwing together a spec for the Scheme-side API that
hides the differences between the database implementation techniques, so
that we'd have an increasingly lucid goal toward which to push our
implementations? I.e. at the level where the user gives a SQL string
with parameters, says to execute it and reads rows, maybe uses
transactions, etc. Also what kind of error representation and error
handling policy is used on the Scheme side.

My subprocess interface is ready to be nudged into whichever direction
the rest of you choose :) I'd like to follow other people's lead on API
design.