Proposal: Remove Num superclass of Bits

Ian Lynagh igloo at earth.li
Sat Oct 15 21:16:26 CEST 2011


On Sat, Oct 15, 2011 at 08:00:21PM +0200, Bas van Dijk wrote:
> 
> We can also combine Gábor's and Roman's solutions to get both
> portability and convenience:
> 
>     zero, one :: a
> 
> #ifdef __GLASGOW_HASKELL__
>     default zero, one :: Num a => a
>     zero = 0
>     one  = 1
> #endif

I think doing this conditionally is a bad idea. Programs will just fail
at runtime when using other impls.

I also think doing it unconditionally is a bad idea. It significantly
raises the barrier to getting another impl to the point where it is
useful.


Hmm, one can be (bit 0) though. I don't think zero is anything nicer
than (clearBit (bit 0) 0). (I don't know what the rules are about using
those functions with arguments > bitSize, so I don't know if that would
be valid for hypothetical bitSize == 0 instances like ()).


Thanks
Ian




More information about the Libraries mailing list