Today, I encountered a strange trouble with higher-rank polymorphism. It was finally solved by nominal typing. Was it a bug in type checker? lack of power in type inference? or lack of my understanding? I'll submit this in hope that it will at least result in better GHC error message some day: My following code https://github.com/sugoi/sugoi/blob/0aec19b29274ddb7c5ae46b9b733e91bae0790b2...
runDB :: Lens NetworkState RIB runDB = lens (f::NetworkState -> RIB) (\x s -> s { _runDB = x }) where f :: NetworkState -> RIB
Didn't typecheck with following error: Sugoi/Types.hs:38:15: Couldn't match expected type `RIB' with actual type `DB.DBMT (Maybe Int) IO a0 -> IO (StM (DB.DBMT (Maybe Int) IO) a0)' Expected type: NetworkState -> RIB Actual type: NetworkState -> DB.DBMT (Maybe Int) IO a0 -> IO (StM (DB.DBMT (Maybe Int) IO) a0) In the first argument of `lens', namely `(f :: NetworkState -> RIB)' In the expression: lens (f :: NetworkState -> RIB) (\ x s -> s {_runDB = x}) How come an expression `(f :: NetworkState -> RIB)' is not of type NetworkState -> RIB ? When I changed the definition of runDB as follows,
runDB = lens (\NetworkState{_runDB = x} -> x) (\x s -> s { _runDB = x })
The source of error was more clear: Expected type: DB.DBMT (Maybe Int) IO a1 -> IO (StM (DB.DBMT (Maybe Int) IO) a1) Actual type: DB.DBMT (Maybe Int) IO a -> IO (StM (DB.DBMT (Maybe Int) IO) a) So, ghc couldn't unify these two types. By using the nominal typing, namely
- type RIB = RunInBase (DB.DBMT (Maybe Int) IO) IO + newtype RIB = RIB (RunInBase (DB.DBMT (Maybe Int) IO) IO)
the problem was solved. Always grateful to haskell-cafe from being there, -- Takayuki MURANUSHI The Hakubi Center for Advanced Research, Kyoto University http://www.hakubi.kyoto-u.ac.jp/02_mem/h22/muranushi.html