[Haskell-cafe] Of types and constructors; a question of programming style

Graham Klyne gk at ninebynine.org
Thu Jul 8 07:23:40 EDT 2004

My original question was related to use of existing facilities.  I find it 
somewhat surprising how rich and complex a language Haskell turns out to 
be.  I'm wary of yet more features.

That said, you seem to be suggesting a kind of "anonymous" union type (to 
complement the proposed "anonymous" record type?).  At one level, that 
seems to answer the question I was raising, and seems to require that 
alternative selection is based on (unification of?) the datatypes, rather 
than by simple matching of constructor labels.  That sounds to me a bit 
like class instance selection (and I note that Scott did suggest using type 

Could your proposal be viewed as a syntactic sugaring of classes and 
instances, without explicit instance declarations?


At 10:35 08/07/04 +0100, Alastair Reid wrote:
>It sounds a little as though you want a family of n-ary union type
>constructor.  A crude first attempt would be:
>   data Either2 t1 t2 = In1 t1 | In2 t2
>   data Either3 t1 t2 t3 = In1 t1 | In2 t2 | In3 t3
>   ...
>but this would be a bit tedious to use because the names are a bit 
>- you'd be relying heavily on type checking to keep you straight.  e.g., it's
>kinds hard to read:
>   f :: Either String Bool -> Either String Bool
>   f (In1 "Fred") = In1 "Frederick"
>   f (In2 m)      = In2 (not m)
>Proposals for records gives you an easy way to define arbitrary product types
>with convenient syntax:
>   { name = "Fred", married = True } :: { name :: String, married :: Bool }
>Maybe what you really want here is a way to define union types with some
>convenient syntax.  In the following made up syntax, |[ ... |] contains a
>list of types to be unioned together with each type labelled by a constructor
>   f :: |[ Name :: String, Married :: Bool |]
>     -> |[ Name :: String, Married :: Bool |]
>   f |[ Name = "Fred" |] = |[ Name = "Frederick |]
>   f [| Married = m |]   = [| Married = not m |]
>I wonder if an extension like this would be generally useful?  (Would it be
>useful for defining a compiler/interpreter by first giving a simple language
>and then adding further operations?)
>Alastair Reid

Graham Klyne
For email:

More information about the Haskell-Cafe mailing list