Martin Sulzmann wrote:
Well, if there's only instance which is not exported, then you can use functional dependencies.
Assume
class C a instance ... => C t
Internally, use
class C a | -> a instance ... => C t
The cases I was looking at had more than one instance, but thats cool! (I didn't realise " -> a" was valid syntax without a LHS for the arrow. Oleg has written quite a bit about using fundeps to close classes. Surely you can export this as well - any attempt to add another instance will conflict with the fundep (-> a) which effectively says there can only be one instance as all the LHS will overlap (all being the empty set)?
Furthermore, there seems to be an issue that has been overlooked so far.
- Providing sufficient type annotations
Well in the toy example, yes... but quite often this occurs where the type is derived and extreemly complex - the whole point is you don't really want to be type annotating every assignment. Also you may really want polymorphism, you just have only one instance at the moment. (duing development, or in a user extensible library) Keean.