| does type-checking remain decidable (in general) for overlapping instances
| (:+o in hugs)?
Type checking in Hugs (with -98, at least) isn't decidable,
either with or without overlapping instances! But decidability
could be recovered by placing stronger syntactic requirements
on the form of class constraints that are allowed on the left
of the = sign in the first line of an instance declarations
(again, either with or without overlapping instances). In other
words, the two things are pretty much independent.
Back in the old days of Gofer, the combination of overlapping
instances with another, more obscure feature of the type system
actually resulted in unsoundness. So far as I know, very few
people ran in to this problem in practice, but it was there.
Hugs doesn't include the second feature (use of "instance
specifics"), so doesn't suffer from the same problem.
Incidentally, the jury is still out, as far as I know, on the
interaction of functional dependencies with overlapping instances.
The theoretical work that I've done on functional dependencies
does not consider the possibility of overlaps. But the implementation
in Hugs does have some experimental extensions (mostly written by
Jeff Lewis) as a first attempt to explore this area.
All the best,
Mark
