[Haskell-cafe] Closed Type Family Simplification

[Richard Eisenberg]

Erik's analysis agrees with mine: type families can never be partially applied, and the fact that Ian's latest example is accepted  as a syntactically correct program is a straightforward bug in GHC. If you try that code on GHC 7.8.2, it will fail.

Type families can *never* be partially applied. (Data families are another story, though...)

Richard

On Aug 14, 2014, at 5:17 AM, Erik Hesselink <hesselink at gmail.com> wrote:

> I think partially applied type synonyms (and type families) are
> supposed to be disallowed, but I recently found out that since GHC
> 7.8, this is not the case anymore, but you get very weird other type
> errors instead. I filed a bug [0], perhaps yours is a similar issue.
> 
> Erik
> 
> [0] https://ghc.haskell.org/trac/ghc/ticket/9433#ticket
> 
> On Thu, Aug 14, 2014 at 5:48 AM, Ian Milligan <ianmllgn at gmail.com> wrote:
>> This is enough to demonstrate this error:
>> 
>> {-# LANGUAGE TypeFamilies #-}
>> {-# LANGUAGE RankNTypes #-}
>> 
>> module TypeFamilyTest where
>> 
>> 
>> type family B f g a where B f g a = f (g a)
>> newtype C f a = C (f a)
>> t :: (forall a. a -> f a) -> (forall a. a -> g a) -> a -> C (B f g) a
>> t f g a = C (f (g a))
>> 
>> Is it not possible to use type families in this way?
>> 
>> 
>> On Wednesday, August 13, 2014 8:39:09 PM UTC-7, Ian Milligan wrote:
>>> 
>>> I suspect the problem happens because I am trying to use a partially
>>> applied type family with a type constructor. However, in trying to replicate
>>> the problem I ran into another strange error message with the following
>>> program (using tagged and constraints as I was trying to keep it as close to
>>> the original program as possible)
>>> 
>>> {-# LANGUAGE TypeFamilies #-}
>>> {-# LANGUAGE ConstraintKinds #-}
>>> {-# LANGUAGE RankNTypes #-}
>>> {-# LANGUAGE ScopedTypeVariables #-}
>>> {-# LANGUAGE TypeOperators #-}
>>> 
>>> module TypeFamilyTest where
>>> 
>>> import GHC.Prim
>>> import Data.Constraint
>>> import Data.Tagged
>>> import Data.Proxy
>>> 
>>> type family A :: * -> Constraint
>>> type family B f g a where B f g a = f (g a)
>>> newtype C f = C (forall a. Tagged a (A a :- A (f a)))
>>> 
>>> t :: forall f g a. C f -> C g -> C (B f g)
>>> t (C x) (C y) = C z where
>>>    z :: forall a. Tagged a (A a :- A (f (g a)))
>>>    z = Tagged (trans (proxy x (Proxy :: Proxy (g a))) (proxy y (Proxy ::
>>> Proxy a)))
>>> 
>>> This fails with the error message
>>> 
>>> TypeFamilyTest.hs:21:19:
>>>    Couldn't match type ‘a1’ with ‘g a1’
>>>      ‘a1’ is a rigid type variable bound by
>>>           a type expected by the context: Tagged a1 (A a1 :- A (f a1))
>>>           at TypeFamilyTest.hs:21:17
>>>    Expected type: Tagged a1 (A a1 :- A (f a1))
>>>      Actual type: Tagged a1 (A a1 :- A (f (g a1)))
>>>    Relevant bindings include
>>>      z :: forall a. Tagged a (A a :- A (f (g a)))
>>>        (bound at TypeFamilyTest.hs:23:5)
>>>      y :: forall a. Tagged a (A a :- A (g a))
>>>        (bound at TypeFamilyTest.hs:21:12)
>>>      t :: C f -> C g -> C (B f g) (bound at TypeFamilyTest.hs:21:1)
>>>    In the first argument of ‘C’, namely ‘z’
>>>    In the expression: C z
>>> 
>>> for some reason C (B f g) expects Tagged a1 (A a1 :- A (f a1)) instead of
>>> Tagged a1 (A a1 :- A (f (g a1)))?
>>> 
>>> On Wednesday, August 13, 2014 6:54:45 PM UTC-7, Richard Eisenberg wrote:
>>>> 
>>>> Your operating assumption sounds right. Do you have a complete, minimal
>>>> example showing the error? If not, I recommend using -fprint-explicit-kinds
>>>> to see if kinds are getting in your way at all.
>>>> 
>>>> Richard
>>>> 
>>>> On Aug 13, 2014, at 8:02 PM, Ian Milligan <ianm... at gmail.com> wrote:
>>>> 
>>>>> When a closed type family has only one instance it seems like it should
>>>>> never fail to simplify. Yet this doesn't appear to be the case. When I
>>>>> defined (in GHC 7.8.3) the closed type family
>>>>> type family (:.:) f g a where (:.:) f g a = f (g a)
>>>>> I get errors such as
>>>>> 'Could not deduce (Object c3 ((:.:) f g a) ~ Object c3 (f (g a)))'
>>>>> (where Object is a Constraint family), indicating that f (g a) is not
>>>>> being substituted for (:.:) f g a as desired. Any idea why this happens?
>>>>> _______________________________________________
>>>>> Haskell-Cafe mailing list
>>>>> Haskel... at haskell.org
>>>>> http://www.haskell.org/mailman/listinfo/haskell-cafe
>>>> 
>>>> _______________________________________________
>>>> Haskell-Cafe mailing list
>>>> Haskel... at haskell.org
>>>> http://www.haskell.org/mailman/listinfo/haskell-cafe
>> 
>> 
>> _______________________________________________
>> Haskell-Cafe mailing list
>> Haskell-Cafe at haskell.org
>> http://www.haskell.org/mailman/listinfo/haskell-cafe
>> 
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>