Proposal: Remove Num superclass of Bits

[wren ng thornton]

On 10/15/11 8:33 PM, Bas van Dijk wrote:
> But an important question is: is it wise to treat booleans and bits
> equally? Because this would allow something like:
>
> {-# LANGUAGE RebindableSyntax #-}
>
> foo = if 1 then 2 else 3

Well, they're both Boolean algebras... But the big question is, or 
should be, what exactly are we trying to model with the Bits class? The 
current design is clearly aimed at modelling bitvectors, and just 
happens to support a few other things too. Given this initial design, 
it's not at all clear that it would be sensible to replace that with a 
model for Boolean algebras.

Indeed, if I were trying to model Boolean algebras, I'd probably start 
with some classes for semilattices and then move up to complemented 
distributive lattices from there. This would set the ground for a better 
handling of partial orders and repairing the inconsistency of 
Float/Double's Ord instances. Unfortunately, it's liable to crash into 
the folks extending Monoid and the like, and it's not entirely clear how 
best to reconcile their interaction.

Personally, I'm -1 for treating Bits like Bool in rebindable syntax. 
While some Bits instances could sensibly be treated as Booleans in 
conditionals, not all of them can. In particular, anything other than 
bivalent Boolean algebras is going to introduce ambiguity. If Bits can 
be Bools, it should be an opt-in system rather than extending 
if_then_else_ to incorporate all Boolean algebras in an unclear and 
ambiguous way.

-- 
Live well,
~wren