Using mutability is a second-optimal solution, and we can have 'ensure-mutable' procedure that copies the input if it's immutable and passes through if it's mutable, as well as the ordinary copy procedure.   Then, an intermediate library that potentially takes functional or linear type arguments can avoid unnecessary copying.  If the caller of such a library wants to pass a mutable shared object, he can just explicitly copy it.

But then, if every intermediate library puts the ensure-mutable procedure, we can just embed it into the original linear-updating API, which became the strategy I outlined in the original email.

In this case, you make implicit what should IMO be explicit. Implicit casting and ad-hoc polymorphism is often not the best idea.