On Sat, 03 Jul 2010 13:28:44 -0700, Dan Doel <[email protected]> wrote:

As a side note, although I agree it abuses the fundeps intent, it was handy
for the specific purpose I was implementing to have a "no-op/passthrough"
instance of op.  In general I like the typedef approach better, but it

                                          ^^^^^^^ should have been "type family", not 
"typedef"

looks like I must sacrifice the no-op to make that switch.
It's potentially not just a violation of intent, but of soundness.

I agree when examining this from the perspective of the compiler.  It's an 
interesting little wormhole in the safety net usually provided by Haskell (or 
more properly GHC in this case) to stop people like me from doing foolish 
things.

From the domain of the original problem, having a no-op instance is still 
desireable, and I achieved this by making the noop instance type polymorphic 
and use the target concrete type to guide the resolution of the interior family 
types.

class C a where
    type A2 a
    type A3 a
    op :: a -> A2 a -> A3 a

data NoOp x = NoOp

instance C  (NoOp b) where
    type A2 (NoOp x) = x
    type A3 (NoOp x) = x
    op _ = id

This is straightforward and less ambiguous and should therefore be safer as 
well.

Thanks and regards,

--
-KQ
_______________________________________________
Haskell-Cafe mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply via email to