Hi Stefan Thanks for a very enlightening reply.
In GHC 6.7.20070712 and Yhc, this is perfectly safe.
In GRIN based systems like Jhc, this is *not* safe, since after evaluation comparisons are done using the full tag.
It's now occurred to me that at a cost of some noise, I could have done things a little differently On 14 Jul 2007, at 20:37, Stefan O'Rear wrote:
On Sat, Jul 14, 2007 at 12:06:30PM +0100, Conor McBride wrote:
A peculiar query for folks who know more about the internals of Haskell compilers than I do. I attach the full code with all the bits and pieces, but let me pull out the essentials in order to state the problem.
newtype Id x = Id x -- element newtype K1 a x = K1 a -- constant
newtype Up1 f p q x = U1 (f (p x) (q x)) type Sum1 = Up1 Either type Prod1 = Up1 (,)
newtype Fst x y = Fst x newtype Snd x y = Snd y newtype K2 a x y = K2 a
newtype Up2 f p q x y = U2 (f (p x y) (q x y)) type Sum2 = Up2 Either type Prod2 = Up2 (,)
class (Bifunctor b, Functor f) => Diag b f | b -> f where diag :: b x x -> f x gaid :: f x -> b x x
and then all of my coercions would (recursively) have been between newtype-isotopes. Would that have more universal traction? I realise, of course that it's all voodoo, and I shouldn't trust a thing. Cheers Conor