Excessive restriction in ghc ?

[Zhanyong Wan]

Hi Sebastien,

Sebastien Carlier wrote:
 
> I am getting an error message from ghc 4.08.1 with
> the following code:
> 
> > class Collection e ce | ce -> e where
> >     empty :: ce
> >     insert :: e -> ce -> ce
> >
> > class (Eq e, Collection e ce) => Set e ce where
> >     member :: e -> ce -> Bool
> >     union :: ce -> ce -> ce
> 
> Main.lhs:7:
>     Class type variable `e' does not appear in method signature
>         union :: {- implicit forall -} ce -> ce -> ce
> 
> Since `ce' uniquely determines `e', I would expect the
> compiler to assume that `e' appears in the method signature.
> Either I am misunderstanding something, or something may be
> missing in the compiler around rename/RnSource.lhs:249.

I encountered the same problem this summer and wrote to Simon PJ and
Jeff Lewis.  Here's Jeff's answer:

> I'm glad to find examples where they are indispensible.  The implementation of
> FDs in GHC is pretty much complete WRT Mark's writeup (but it doesn't complain
> about instances inconsistent with FDs).  I'm using them in a current project,
> but in a fairly conservative manner.  In hugs, I implemented several
> extensions to do with derived instances and superclasses - pretty much
> necessary as you've found.  Unfortunately, in hugs I implemented it in rather
> the wrong way.  Based on dicsussions at the Hugs/GHC meeting w/ Simon, I have
> a cunning plan for finishing the implementation properly in GHC, but just
> haven't had the chance to do it.  What I need to do is write it up, so that
> either Simon or myself can finish the job.

So the short answer to your question is: FD in derived instances is not
implemented in GHC yet.

I'm still eagerly waiting to use this feature in my project.  Jeff,
could you give us an update on the progress?  Thanks!

-- 
# Zhanyong Wan     http://pantheon.yale.edu/~zw23/ ____
# Yale University, Dept of Computer Science       /\___\
# P.O.Box 208285, New Haven, CT 06520-8285        ||___|