Sebastian Graf
OK, so just deriving the instances doesn't yield the expected behavior, because the `Var` case explicitly mentions `Id`s instead of the type parameter `b`. Even if that would be changed, it's not easy to pin down over which parts of the syntax tree we should 'map'. Should we include binding sites of local variables? I'm inclined to say No, but only because I have my concrete use case in mind.
It's probably best to have non-derived, non-typeclass functions `foldMapVars :: Monoid m => (Id -> m) -> Expr b -> m`, or a variant where the mapping function also gets supplied a value of `data NameSite = Lam | Let | VarRef` (for lack of a better name).
Agreed, I think there is enough subtlety here that it's best to be explicit about which things in particular you want to traverse. Cheers, - Ben