GitLab

This self-contained module:

{-# LANGUAGE FlexibleInstances, OverlappingInstances, UndecidableInstances #-}

module Foo where

class Sat a
class Data ctx a
instance  Sat (ctx Char)             => Data ctx Char
instance (Sat (ctx [a]), Data ctx a) => Data ctx [a]

class Data FooD a => Foo a

data FooD a = FooD

instance Foo t => Sat (FooD t)

instance Data FooD a => Foo a
instance Foo a       => Foo [a]
instance                Foo [Char]

gives this error in the HEAD (and a similar, but less helpful, error in 6.6):

$ ghc -c Foo.hs

Foo.hs:18:0:
    Overlapping instances for Foo [a]
      arising from the superclasses of an instance declaration
                   at Foo.hs:18:0
    Matching instances:
      instance [overlap ok] (Foo a) => Foo [a]
        -- Defined at Foo.hs:18:0-30
      instance [overlap ok] Foo [Char] -- Defined at Foo.hs:19:0-33
    (The choice depends on the instantiation of `a'
     Use -fallow-incoherent-instances to use the first choice above)
    In the instance declaration for `Foo [a]'

Simon writes:

The reason is this:

In the instance for Foo [a]
we need a superclass (Data FooD [a])
using the instance for Data, we therefore need
        (Sat (FooD [a], Data FooD a)
But we are given Foo a, and hence its superclass Data FooD a
So that leaves (Sat (FooD [a]))
Using the instance for Sat means
we need (Foo [a])

At that point GHC says there are two instances that match Foo [a], and which one
to choose depends on how you instantiate a.

What we wanted was to "tie the knot" and use Foo [a] recursively, the very one
we are constructing.  GHC does that when solving Foo [a], but not when
constructing it.

This is a good point, I will think about it.  It's a bit delicate.  Meanwhile,
can you Trac it?  It's a nice example.

A larger module showing how the instances above can be desirable in a real program is this (requires the syb-with-class package from hackage):

{-# LANGUAGE TemplateHaskell, FlexibleInstances, CPP,
             OverlappingInstances, UndecidableInstances #-}

module Main where

import Data.Generics.SYB.WithClass.Basics
import Data.Generics.SYB.WithClass.Instances ()

data Element = Elem String [Element]
             | CData String
    deriving Show

class Data XmlD a => Xml a where
    toXml :: a -> [Element]
    toXml = defaultToXml

data XmlD a = XmlD { toXmlD :: a -> [Element] }

xmlProxy :: Proxy XmlD
xmlProxy = error "xmlProxy"

instance Xml t => Sat (XmlD t) where
    dict = XmlD { toXmlD = toXml }

defaultToXml :: Xml t => t -> [Element]
defaultToXml x = [Elem (constring $ toConstr xmlProxy x)
                       (concat $ gmapQ xmlProxy (toXmlD dict) x)]

-----

instance Data XmlD t => Xml t

#ifdef STRING
instance Xml String where
    toXml x = [Elem "string" [CData x]]
#endif

#ifdef LIST
instance Xml a => Xml [a] where
    toXml = concatMap toXml
#endif

main :: IO ()
main = do print $ toXml (Just True)
          print $ toXml [True, False]
          print $ toXml "foo"

which allows us to specialise XML marshalling for either lists (ignoring all the (:) and [] constructors) or strings (turn them into CData), but not both:

$ ghci -v0 -DLIST z.hs
*Main> main
[Elem "Just" [Elem "True" []]]
[Elem "True" [],Elem "False" []]
[Elem "f" [],Elem "o" [],Elem "o" []]
*Main>
$ ghci -v0 -DSTRING z.hs
*Main> main
[Elem "Just" [Elem "True" []]]
[Elem "(:)" [Elem "True" [],Elem "(:)" [Elem "False" [],Elem "[]" []]]]
[Elem "string" [CData "foo"]]
*Main>
$ ghci -v0 -DLIST -DSTRING z.hs
[overlapping instances error]

If we replace the

instance Xml a => Xml [a] where

line with

instance (Xml a, Xml [a]) => Xml [a] where

then everything seems to work out, but this feels a little dodgy.

Trac metadata

Trac field	Value
Version	6.6.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler (Type checker)
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system	Unknown
Architecture	Unknown