The mapping mechanism looks good for the schema structure. But I cannot vote in favor of this package (nor against completely), because it is quite naive when it comes to handling the SQL identifiers.
In principle, the SQL specs do not limit what you can name a schema, database, table or field. If you want to, you could name a field:
including the spaces there, even use any charset (accents, etc.)
That is why I pointed out the ISO spec for doing just this mapping when this idea was first discussed in the pear-dev list. There is even a freely downloadable draft you could've used.
I know the ISO standard is big and boring (I did an implementation in Java of this a couple of years ago in my previous job), and it should be considered for a future release or even before the first release. That way it would be applicable in a more general case. Even more in view of the fact that some of the big db vendors support importing from the ISO standard XML representation into their own dbs (Oracle and DB2 come to mind).