On 10/1/2014 11:42 PM, Bardur Arantsson wrote:
(I say "proposal"... this needs some serious fleshing out and a semi-formal specification for instance resolution rules &c before it becomes a proposal. Just "Look at Oleg's paper" is not good enough.) I see I came off unclearly. Please allow me to remedy that.
My reference to Oleg's paper was badly framed. What I meant to say was more along the following lines: Oleg discussed adding local typeclass instances in his reflection paper. For some reason, his proposal wasn't taken. (Does anyone know why?) I'd like to suggest something along similar lines: Allow instances in let-expressions and where-clauses. These instances shadow other instances with the same parameter list. Similarly, instances defined in modules shadow imported instances. Thus, in the following program:
module Foo where import Bar (Baz) -- Where Bar defines instance Show (Baz a) instance Show (Baz a) where -- shadows the imported instance foo = let instance Show (Baz a) where -- shadows the instance defined in ... -- in where below, just like other bindings where instance Show (Baz a) -- shadows the module-global instance instance Show (Baz Quz) -- only allowed with -- OverlappingInstances, isn't shadowed by any of the other -- instances in this example.
In this example, the instances for Baz have the following scopes: * From Bar: Nowhere, as it is immediately shadowed. * From Foo: Within all top-level declarations besides foo, as foo's where clause shadows it. * From the where clause: Within the where clause. The instance for Baz Quz is also visible within the let expression, as it is not shadowed by it. * From the let expression: Within the let expression. Basically all I'm doing is applying regular name resolution semantics to instances with the same parameter lists. I hope this helps clarify things. If not, please tell me what needs to be done to fully clarify this. Gesh