Generalized newtype deriving 6.6 vs. 6.8
Hello, I noticed there is a difference in generalized newtype deriving between 6.6 and 6.8. In GHC 6.4.1 the following:
{-# LANGUAGE GeneralizedNewtypeDeriving #-}
import Control.Monad.Error
newtype MyMonad m a = MyMonad (m a) deriving (Monad, MonadError String)
correctly gives a derived instance MonadError String m => MonadError String (MyMonad m) The new GHC 6.8.1 complains with: No instance for (MonadError String m) arising from the 'deriving' clause of a data type declaration at DeriveTest.hs:(5,0)-(6,36) Possible fix: add an instance declaration for (MonadError String m) When deriving the instance for (MonadError String (MyMonad m)) Generalizing the instance to MonadError e m => MonadError e (MyMonad m) works in both versions. Twan
Well it's debatable. Suppose we have newtype Foo = MkFoo String deriving( Num ) Do you want to generate instance Num String => Num Foo ? I suspect not -- usually we generate an error message right away if we need a Num String instance and one is not available. Now you could argue that | MonadError String m => MonadError String (MyMonad m) is more plausible because the 'm' is somehow the important bit. But it gets into undecidable-instance territory, and at that point you (now) just have to write the instance declaration yourself. The manual is misleading on this point. It says "for each ci, the derived instance declaration is: instance ci t => ci (T v1...vk)". But actually GHC tries to figure out the context, just as for other derived instance decls, and insists (for the reasons above) that it reduces to type variables only. Does that make sense? I'll update the documentation if so. Simon -----Original Message----- | From: glasgow-haskell-users-bounces@haskell.org [mailto:glasgow-haskell-users-bounces@haskell.org] On Behalf Of | Twan van Laarhoven | Sent: 03 November 2007 18:08 | To: glasgow-haskell-users@haskell.org | Subject: Generalized newtype deriving 6.6 vs. 6.8 | | Hello, | | I noticed there is a difference in generalized newtype deriving between | 6.6 and 6.8. In GHC 6.4.1 the following: | | > {-# LANGUAGE GeneralizedNewtypeDeriving #-} | > | > import Control.Monad.Error | > | > newtype MyMonad m a = MyMonad (m a) | > deriving (Monad, MonadError String) | | correctly gives a derived instance | | MonadError String m => MonadError String (MyMonad m) | | The new GHC 6.8.1 complains with: | | No instance for (MonadError String m) | arising from the 'deriving' clause of a data type declaration | at DeriveTest.hs:(5,0)-(6,36) | Possible fix: add an instance declaration for (MonadError String m) | When deriving the instance for (MonadError String (MyMonad m)) | | Generalizing the instance to | | MonadError e m => MonadError e (MyMonad m) | | works in both versions. | | Twan | _______________________________________________ | Glasgow-haskell-users mailing list | Glasgow-haskell-users@haskell.org | http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
Simon Peyton-Jones wrote:
Well it's debatable. Suppose we have
newtype Foo = MkFoo String deriving( Num )
Do you want to generate
instance Num String => Num Foo
?
Personally, I think I would like that -- along with a warning message. It makes it clearer to me that I can do newtype Foo a = MkFoo a deriving( Num ) without requiring that Foo can only be instantiated with types in Num. I'm not sure though. Isaac
participants (3)
-
Isaac Dupree -
Simon Peyton-Jones -
Twan van Laarhoven