That is a plausible scheme,

That is a plausible scheme, but probably not very attractive for an FPL. Phew! I thought I was taking crazy pills for awhile there! :-) Instead, you associate information with every type variable, to provide what is not known statically. Right, isn't that what I'm doing though? I think the representation I described is only needed for polymorphic function parameters. Other parameters are known fully at compile-time as you say. Right, isn't that what I'm doing though? From what I understood, you extend the representation of structured objects with a descriptor pointer. That seems unavoidable in this approach. That pointer is tied to values, not type parameters. And since all objects can be used in monomorphic as well as polymorphic contexts, you cannot simply equip them selectively. IOW, there is no such thing as a... [read full story]