rae at cs.brynmawr.edu
Fri Oct 7 01:00:56 UTC 2016
> On Oct 6, 2016, at 5:36 PM, Ben Gamari <ben at smart-cactus.org> wrote:
> The motivation here isn't that we want to decompose this. It's that
> carrying the kind variable instantiations will shrink the size of your
> typical (non-kind-polymorphic) TypeRep and may actually simplify a few
> things in the implementation. Namely, I suspect that the trouble we
> currently experience due to recursive kind relationships will vanish.
Ah. This is great motivation.
>> Why not carry both the kind instantiations and the
>> final kind? That seems much simpler than the scheme below.
> Admittedly there is a bit of complexity here. I haven't yet tried
> implementing it but I suspect it's manageable.
On further thought, my idea doesn't simplify any of the TyConKindRep stuff, so I retract my idea.
> Oh dear. This will require some thought.
Indeed. It's never easy. :)
More information about the ghc-devs