typechecker loops on simple program with fundep

[guest]

The program below causes the "Renamer/typechecker" phase to loop. Confirmed with GHC 6.8.20070927 on Linux and 6.9.20071018 on Windows (so I'm not absolutely certain it's present in latest 6.8, but it seems very likely).

The program has an infinite type - the definition of translate causes (String, a) to be unified with a. Removing the fundep in MkA causes a reasonable error to be reported.

{-# LANGUAGE MultiParamTypeClasses, FunctionalDependencies, FlexibleInstances #-}
module X() where

data A a = A

class MkA a b | a -> b where
   mkA :: a -> A b

instance MkA a a where

translate :: (String, a) -> A a
translate a = mkA a

Ganesh

PS: See also #1797 and #1781

[igloo]

This also happens with my 6.8.0.20071019, and is a regression since 6.6.

[simonpj]

Definitely a bug. It happens because the improvement gives rise to an equality (a ~ (String,a)), and this equality is simply generated but not discharged, by TcSimplify.tcImprove.

Simon

[simonpj]

Fixed! The error message is different than before:

FD3.hs:22:14:
    Could not deduce (MkA (String, a) a) from the context ()
      arising from a use of `mkA' at FD3.hs:22:14-18

but it's still reasonable.

Patch is this one:

Sat Oct 27 16:54:59 GMT Daylight Time 2007  simonpj@microsoft.com
  * Make 'improvement' work properly in TcSimplify

Thanks Ganesh!