[Simon mentioned my work on `functional dependencies' in one of his
messages a couple of days ago, so I thought I'd better post an
explanation!]
A couple of months ago, I developed and implemented an extension to
Hugs that has the potential to make multiple parameter type classes
more useful. The key idea is to allow class declarations to be
annoted with `functional dependencies'---an idea that has previously
received rather more attention in the theory of relational databases.
For example, we could use the following declaration as the starting
point for a simple `collection' class library:
class Collects e ce | ce -> e where
empty :: ce
insert :: e -> ce -> ce
member :: e -> ce -> Bool
The new part here is the clause "| ce -> e", which tells us that the
"ce" parameter (the type of the collection) uniquely determines the
"e" parameter (the type of the elements). Dependencies like this
can be used to avoid ambiguity problems; to obtain more accurate, and
less cluttered inferred types; and to allow more general sets of
instances than constructor classes. If this has got your interest,
you can find more information at http://www.cse.ogi.edu/~mpj/fds.html.
Support for this extension is included in the imminent September 1999
release of Hugs 98. In the meantime, Jeff Lewis, a colleague here at
OGI, has been putting together an implementation for GHC. I've also
written a longer paper that gives more technical details, and explains
how this idea fits in with other work, but there are a few things I
want to do to that before it is ready for distribution, probably after
I get back from the Haskell Workshop.
I hope folks will find this useful. A month or two back, somebody on
this list said that we had no `peer review' process ... people
just do something, then say "here it is, use it". Well I've done
something ... here it is ... I hope that some of you will use it ...
and I do very much appreciate your feedback. The peer review
starts today!
All the best,
Mark
