[tree-ssa] Death to WFL nodes

[Andrew MacLeod]

> Another would be to have a void * pointer in the tree_common structure.  
> Initially it would hold the address of the locus information.  When we
> move into the tree-ssa code (or whenever you need your annotations) the
> locus pointer is moved into your tree annotation and the tree common void *
> pointer points to the annotation.  We'd do the reverse once we're done with
> tree-ssa opts. This is more memory efficient, but certainly uglier.
> 

eeww, lets not do this :-) Then any generic routines which access this
info won't work on SSA annotated trees. Let SSA and its annotations have
its own overhead instead of trying to salvage memory this way.  We're
also asking for propagation errors when we duplicate or manipulate code
within the SSA framework, because now the annotations have information
which is not SSA related. 

I think the can of worms there is a bit too large.

And as you say, its uuuuuuuugggggggly.  eww eww eww.

Andrew