Tom, Yes of course it is simplified for clarity. Here is a modified version where fb does something (a bit more usefull) class PP m where create :: a -> m a data A a = A a instance PP A where create a = A a class CB a where fb :: a -> a data B m a = B (m a) instance (PP m) => PP (B m) where create a = let a' = fb a in B (create a') class PP' m a where create' :: a -> m a instance (PP m) => PP' m a where create' = create instance (PP m, CB a) => PP' (B m) a where create' a = let a' = fb a in B (create a') Actually I ran into that problem when trying to add a kind of rule engine layer above the Persistent typeclass. Given the complexity of these typeclass, I think it is more practical to reason about a simpler form of the same problem. Thanks J-C On Tue, Apr 8, 2014 at 9:42 AM, Tom Ellis < tom-lists-haskell-cafe-2013@jaguarpaw.co.uk> wrote:

On Tue, Apr 08, 2014 at 09:25:17AM +0200, jean-christophe mincke wrote:

...

instance (PP m) => PP (B m) where create a = let _ = fb a in B (create a)

Your use of 'fb' here is baffling. Am I right in thinking you have tried to simplify your problem for clarity? If so I think you have simplified too far!

Could you give an example where the use of 'fb' actually matters?

Tom _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

I've run into problems like this as well. Often you can redesign a bit to fix things, although it depends on the application how to do that. Two other technical options in addition to your PP': You can add a constraint to the 'create' function. It depends on the actual semantics if this makes sense, I guess: class PP2 m where create :: CB a => a -> m a You can use ConstraintKinds [1] to add a constraint to create depending on the type 'm': type family C m a :: Constraint class PP3 m where create :: C m a => a -> m a This way you can choose if you want to add a constraint and which one, depending on the 'm'. If you can solve it without any of these things, I'd probably prefer that personally, though. As Tom also said, complicated type class constructions often make your program more difficult to understand and maintain. Erik [1] http://www.haskell.org/ghc/docs/7.4.1/html/users_guide/constraint-kind.html On Tue, Apr 8, 2014 at 10:17 AM, jean-christophe mincke wrote:

Tom,

Yes of course it is simplified for clarity.

Here is a modified version where fb does something (a bit more usefull)

class PP m where create :: a -> m a

data A a = A a instance PP A where create a = A a

class CB a where fb :: a -> a

data B m a = B (m a) instance (PP m) => PP (B m) where create a = let a' = fb a in B (create a')

class PP' m a where create' :: a -> m a

instance (PP m) => PP' m a where create' = create

instance (PP m, CB a) => PP' (B m) a where create' a = let a' = fb a in B (create a')

Actually I ran into that problem when trying to add a kind of rule engine layer above the Persistent typeclass. Given the complexity of these typeclass, I think it is more practical to reason about a simpler form of the same problem.

Thanks

J-C

On Tue, Apr 8, 2014 at 9:42 AM, Tom Ellis wrote:

...

On Tue, Apr 08, 2014 at 09:25:17AM +0200, jean-christophe mincke wrote:

...

instance (PP m) => PP (B m) where create a = let _ = fb a in B (create a)

Your use of 'fb' here is baffling. Am I right in thinking you have tried to simplify your problem for clarity? If so I think you have simplified too far!

Could you give an example where the use of 'fb' actually matters?

Tom _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe