On Friday 11 February 2011 19:33:27, Cristiano Paris wrote: > Hi cafè, > > given the following toy code: > > --------------- > module Main where > > class Dumb p where > dumb :: p -> String > > newtype Branded a b = Branded b > > unbrand :: Branded a b -> b > unbrand (Branded x) = x > > wrong :: Dumb a => b -> Branded a b > wrong = Branded > > right :: Show a => b -> Branded a b > right = Branded > --------------- > > Why: > > --------------- > quarry:Haskell paris$ ghci -O1 > GHCi, version 6.12.3: http://www.haskell.org/ghc/ :? for help > *Main> unbrand $ right True > True > *Main> unbrand $ right "Foo" > "Foo" > --------------- > > but: > > --------------- > *Main> unbrand $ wrong True > > <interactive>:1:10: > Ambiguous type variable `a' in the constraint: > `Dumb a' arising from a use of `wrong' at <interactive>:1:10-19 > Probable fix: add a type signature that fixes these type variable(s) > --------------- > > ? > > Maybe it's a dumb question but... thank you for any explanation...
It's because there's no way to determine the type variable a (in either wrong or right). In such cases, under some circumstances, the type variable gets defaulted (the ambiguous contexts all must have the form (C a), at least one numeric class [Num, Integral, Fractional, ...] is involved, all involved classes come from the Prelude or the standard libraries) as specified in the report (section 4.3, iirc). These conditions are not met by your 'right' function, but ghci uses extended default rules, a type variable a with a Show constraint [and no other] gets defaulted to (). But ghci has no rule how to default your Dumb class, so it reports the ambiguous type variable there. [Without extended defaulting rules, ambiguous type variable errors would be too frequent at the ghci prompt.] If you try to compile the module, both should raise an ambiguous type variable error, since the compiler follows the standard defaulting rules (unless you enable -XExtendedDefaultRules). _______________________________________________ Haskell-Cafe mailing list [email protected] http://www.haskell.org/mailman/listinfo/haskell-cafe
