I meant to make this last point myself, Tom, and forgot. NULL and the empty string are both passed back as empty Tcl strings in the ns_set built by the drivers (ours and the sybase one written by the AOLserver team - actually, the PG driver was originally written by them, too, though bindvar emulation was added by Dan Wickstrom and me).
So there's symmetry here.
Dave ... the db_null idea's a noble failure, unfortunately, because it doesn't work with either real bindvars or bindvar emulation. Or rather the only value that will work is the empty string, for any RDBMS we use, so it's pointless.
Stripping it out and all references to it is one of those very low-priority things on my personal list for us to clean up someday.
