Alex Baretta <alex <at> barettadeit.com> writes:
> 
> Richard Jones wrote:
> >   [I just throwing this idea out there to see if people find it
> >   interesting, or want to shoot it down ...  There're only fragments of
> >   working code at the moment]
> > 
> > I'm thinking about a type safe interface to PostgreSQL.  One such
> > interface at the moment is in Xcaml, but it only supports a very small
> > subset of SQL, and I like to use complex SQL.  It seems that there is
> > a way to support the whole of PostgreSQL's language from within OCaml
> > programs, in a type safe way.
> 
> Every once in a while we extend the Embedded SQL with a new feature, but
> we never planned to support all of PostgreSQL. In fact, what we want to
> have is abstraction over the actual DB implementation.
> 
> > The general plan would be to have a camlp4 extension which would use
> > Postgres's new "PREPARE" feature to actually prepare the statements,
> > and Postgres's other new feature, "Describe Statement", to pull out
> > the parameter types and result types from the prepared statement.
> > This allows the camlp4 extension to replace the statement string with
> > a type safe expression, and allow type inference to find mismatches.
> > How a typical program would look is shown at the end of this message.
> 

Back in the pre-internet era, Dec implemented a DBMS (Rdb I think), a C++
compiler and a "compile time" coherency check between C++ and Rdb.

Their implementation had the following features
-1) C++ compilation would read Rdb schema,
-2) There was a strong coupling between database schema and C++ program.
-3) Moving from test environment to production lead us to rebuild the code,
(That’s silly but I could not find a way around it).

We just gave up using it.


> I really think XDBS is the the way to go. You define the schema in a
> high level language (OO-Entity-Relationship modeling), supporting lower
> level refinements (logical, physical and virtual schema refinements) and
> compiling to Ocaml and SQL-DDL. This way, the type safety can be
> established at compile time without need for a database connection.
> Also, the type safety does not depend on a specific implementation of
> SQL, which is usually desireable.
> 
> Alex
> 

Nb: I have never worked with PostgreSQL, BUT, with Oracle, Informix, DB2,
sybase, mssql, the full name of a table ( databasename.login.table ) only binds
to an entry inside a system catalog.

If your application uses several logins, then, checking program structures
against database schema can not really occur before login time.

If the goal is only a sanity check, then using any reference definition can
help, but I don't think it will replase the run time check.

Regards,