[PATCH][RFC] Step 1 to a middle-end type-system

[Richard Kenner]

> Note that my particular suggestion was that #2 be accomplished by
> adding a field to the types (TYPE_EQUIVALENCE), rather than
> transforming the actual types.   If you transform the actual types,
> you have to do something about debug info.

I don't see the issue as transforming types, but of deciding which NOP_EXPRs
can be safely deleted and when two types are equivalent for the purpose
of validating the tree.  Debugging info would not be affected since you
wouldn't be touching the TREE_TYPE of a decl.