Is there already a list class?

[Robert Dockins]

On Jan 9, 2007, at 9:12 PM, Spencer Janssen wrote:

> On Jan 9, 2007, at 7:52 PM, Marc Weber wrote:
>> Hello.
>>
>> Is here the right place to request a list class?
>> eg
>> class List l e where
>> 	(:) :: e -> l e -> l e
>> 	head :: ..
>
> Note that this approach isn't quite flexible enough.  Your example  
> forces the container type to have kind * -> *, and therefore can't  
> support certain specialized containers like ByteString.
>
> There is a class like this in the Edison library (http:// 
> www.eecs.tufts.edu/~rdocki01/edison.html), it is called Seq  
> (Haddocks: http://www.eecs.tufts.edu/~rdocki01/docs/edison/Data- 
> Edison-Seq.html).  However, it suffers the same kind flexibility  
> issues as your List class.  Other classes in Edison take a MPTC 
> +fundep approach and I'm not sure why Seq doesn't.  Can you comment  
> on this, Rob?

I consider this mostly a historical artifact.  The Edison design  
dates back to about 1998, before fundeps were available in a Haskell  
implementation.  The Set/Bag and Finite Map classes were designed  
with MPTC, but no fundeps.  The sequence class was nicer because it  
was more elegant, and played nicer with type inference.  In the  
course of time, Set/Bag and Finite Map classes got fundeps, and  
became a bit nicer.  When I took over maintenance, the typeclass  
hierarchy was much as it is now.

I am personally in favor of the idea of changing the sequence class  
to the MPTC+fundep approach, for largely the reasons you've  
mentioned.  The downsides are twofold: 1) Functor, Monad, and  
MonadPlus could no longer be superclasses of Sequence and 2) its a  
pretty major API change.  Despite the downsides, I've become  
convinced this is the right direction to go, and this change will  
almost certainly take place sometime in the not too distant future.   
I'm now finished with some work which was occupying most my attention  
for the last six months or so.


I am also considering a more sweeping API reorganization, where the  
typeclasses become less monolithic (especially the sequence class),  
and the non-observable classes go away, but I'm still trying to  
figure out the most optimal way to restructure things, so this change  
won't happen for awhile.


>> This might be used in Data.Set, Data.Map
>>
>> class StorableAsList l e t where
>> 	fromList :: l e ->  t
>> 	toList :: t -> l e
>
> This is subsumed by other Edison functionality.
>
>> I'd like to help implementing/ writing it.
>>
>> Do you consider this beeing a useful enhancement?
>
> Oh yes, but let's avoid reinventing the wheel if at all possible.
>
>
> Spencer Janssen


Rob Dockins

Speak softly and drive a Sherman tank.
Laugh hard; it's a long way to the bank.
           -- TMBG