Cryptic type error message (should be syntax error)

[josef.svenningsson@…]

It seems ghc allows a syntax for types which if beyond Haskell98 even without -fglasgow-exts. Consider the following program:

destroy :: (forall a. (a -> Maybe (b,a)) -> a -> c) -> [b] -> c
destroy g xs = g listpsi xs
  where listpsi :: [a] -> Maybe (a,[a])
        listpsi [] = Nothing
        listpsi (x:xs) = Just (x,xs)

It gives the following rather cryptic error message without -fglasgow-exts:

ConcatMap.hs:7:17:
    Couldn't match expected type `.' (a rigid variable)
           against inferred type `(->)'
      `.' is bound by the type signature for `destroy'
        at ConcatMap.hs:6:20
    Probable cause: `listpsi' is applied to too few arguments
    In the first argument of `g', namely `listpsi'
    In the expression: g listpsi xs

The fact that ghc interpreted `.' as a type variable baffeled me for a moment. I would expect ghc to report a syntax error since the type is not syntactically Haskell98.

[igloo]

Haskell 98 indeed does not seem to allow operators as types, so we ought not allow this without -fglasgow-exts. Too late for 6.6, though...

[simonpj]

Fixed thank you. Test is tcfail166

Simon

[simonmar]

Definitely fixed, error is now:

919.hs:1:20:
    Illegal operator `.' in type `forall a . ((a -> Maybe (b, a))
                                              -> a -> c)'
      (Use -XTypeOperators to allow operators in types)