On Tue, 17 Apr 2007, Bertram Felgenhauer wrote:
Henning Thielemann wrote:
I wonder whether the Ord instance is a good choice as constraint for Data.Map, Data.Set and so on.
In the case where there are Ord instances they are a good choice. Adding another type class for this purpose has the disadvantage that you have to write class instances for them, for all existing types that have Ord instances; while trivial this still adds up to quite a few lines of boilerplate code. This has to be weighted against the added boilerplate code for cases where a total order can be defined but is rather artificial. Right now I think that this is the exception rather than the rule. Changing the library interface should also not be done lightly.
I always think in terms of "How would it be done correctly if we would start from scratch?" :-)
Personally I'd do it without an Indexable class; just
newtype Index a = Index a
instance Ord x => Ord (Index (Complex x)) where (Index (Complex a b)) `compare` (Index (Complex c d)) = (a, b) `compare` (c, d)
should be good enough. 'Index' marks the purpose for the type.
That requires extended instance declarations, doesn't it?