less cpp noise in properties using type synonyms

[chessai]

instead of:

prop1 :: forall proxy f.
#if HAVE_QUANTIFIED_CONSTRAINTS
(blah)
#else 
(blah1)
#endif 
  => proxy f -> Property
...

prop1 :: forall proxy f.
#if HAVE_QUANTIFIED_CONSTRAINTS
(blah)
#else 
(blah1)
#endif 
  => proxy f -> Property

we can have, using Bifunctor as an example, the following:

type BifunctorProp f =
  ( Bifunctor f
#if HAVE_QUANTIFIED_CONSTRAINTS
  , forall x y. (Eq x, Eq y) => Eq (f x y)
  , forall x y. (Show x, Show y) => Show (f x y)
  , forall x y. (Arbitrary x, Arbitrary y) => Arbitrary (f x y) 
#else 
  , Eq2 f
  , Show2 f
  , Arbitrary2 f
#endif 
  ) => proxy f -> Property

prop1 :: forall f. BifunctorProp f
prop2 :: forall f. BifunctorProp f
...
prop3million :: forall f. BifunctorProp f

this would help reduce at least some of the noise, and wouldn't be visible externally, since it only affects the internal properties (ie not the user-facing functions of type Proxy a -> Laws)

[chessai]

this requires constraintkinds. do those work in 7.4? a comment from ekmett in the documentation for the constraints package seems to indicate that they do not.

[andrewthad]

I think this is a good idea. Even without ConstraintKinds:

class 
  ( Bifunctor f
#if HAVE_QUANTIFIED_CONSTRAINTS
  , forall x y. (Eq x, Eq y) => Eq (f x y)
  , forall x y. (Show x, Show y) => Show (f x y)
  , forall x y. (Arbitrary x, Arbitrary y) => Arbitrary (f x y) 
#else 
  , Eq2 f
  , Show2 f
  , Arbitrary2 f
#endif
  ) => TestableBifunctor f

instance 
  ( Bifunctor f
#if HAVE_QUANTIFIED_CONSTRAINTS
  , forall x y. (Eq x, Eq y) => Eq (f x y)
  , forall x y. (Show x, Show y) => Show (f x y)
  , forall x y. (Arbitrary x, Arbitrary y) => Arbitrary (f x y) 
#else 
  , Eq2 f
  , Show2 f
  , Arbitrary2 f
#endif
  ) => TestableBifunctor f

prop1 :: forall f. TestableBifunctor f => proxy f -> Property
prop2 :: forall f. TestableBifunctor f => proxy f -> Property

In the values exposed to the end user, we would still want to make sure that we write out the full signature.

[chessai]

seems a waste to introduce a new class. pollution. i would rather just wait until we don't have to support 7.4, since 7.6 is when ConstraintKinds don't crash GHC.

[andrewthad]

The class wouldn't be visible to users, but regardless I'm fine with waiting until primitive drops support for GHC 7.4.