While I agree the spicing the CSV driver might be a good starting point, one still has to remeber that just a plain CSV file won't cut it for a few issues. Namely:

• filesystem performance
• datatype support (longs, lobs, raws, etc)
• field constraints
• table constraints
• roles and permissions
• common objects (databases, tablespaces, etc)
• client server support

I guess what I'm getting at is: wouldn't an entirely new data format be needed to support all of the above features, or could these be acheived in some mischeivious (spellchecker needed) "linking-globing-OS glueing" kinda way?

This question has definitly become a meditation...