instead of to have a chained series of scope nodes.  This makes
the generated table smaller, improves the efficiency of the
interpreter, and make the factoring optimization much more 
reasonable to implement.

llvm-svn: 97160