On 7 September 2010 02:53, John Lato
I'd like to make one more argument in favor of my preference for more splitting of type classes. IMO it's beneficial to split up classes to minimize unnecessary dependencies. That is, while e.g. Monoid is very useful for containers, many container methods won't need it, e.g. "elem" or "filter". When you divide the type classes so only what's necessary is required, more data structures will be able to meet those criteria, which in turn makes the class more generally useful.
Well, my current work is to implement restricted versions of the various type classes (I'm undecided if they should be split off into a separate package or not) and then have the Data.Containers module basically define type aliases (which will probably still use Monoid though, since the vast majority of containers are able to be empty).
As an example, I have an extension to iteratee which uses raw buffers for holding data. This allows it to operate in truly constant memory (only one allocation, deallocated when it's finished), which has resulted in performance benefits. Unfortunately there is not a good Monoid instance for this type because appending buffers is not possible. By introducing the NullPoint class (equivalent to mempty of Monoid) in iteratee, my raw buffer extension is able to re-use several iteratee functions which it would not be able to if I had require Monoid instead of NullPoint.
Hmmm, splitting up Monoid might make sense. I'll have a think about that (I was going to just keep Monoid as is since no constraints are needed. -- Ivan Lazar Miljenovic Ivan.Miljenovic@gmail.com IvanMiljenovic.wordpress.com