#12406: New not-in-scope behaviour + deferred typed holes = disappearing error messages -------------------------------------+------------------------------------- Reporter: ertes | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Description changed by ertes: @@ -1,2 +1,1 @@ - The following code snippet (using the [https://hackage.haskell.org/package - /ref-tf ref-tf library]) is well-typed, and GHC is fine with it: + The following code snippet is well-typed, and GHC is fine with it: @@ -5,0 +4,2 @@ + {-# LANGUAGE TypeFamilies #-} + @@ -6,1 +7,11 @@ - import Control.Monad.Ref + import Data.IORef + + class MonadRef m where + type Ref m :: * -> * + newRef :: a -> m (Ref m a) + readRef :: Ref m a -> m a + + instance MonadRef IO where + type Ref IO = IORef + newRef = newIORef + readRef = readIORef @@ -12,2 +23,3 @@ - However, if one comments out the first line, then GHC treats `join` as a - typed hole due to #10569, but fails to infer its type, causing errors: + However, if one removes the import of `Control.Monad`, then GHC treats + `join` as a typed hole due to #10569, but fails to infer its type, causing + errors: @@ -16,2 +28,2 @@ - test.hs:10:8-23: error: … - • Couldn't match type ‘Ref m0’ with ‘GHC.IORef.IORef’ + test.hs:17:8-23: error: … + • Couldn't match type ‘Ref m0’ with ‘IORef’ @@ -19,1 +31,1 @@ - test.hs:10:29-32: error: … + test.hs:17:29-32: error: … @@ -24,2 +36,2 @@ - holes` is in effect, only the type error is reported leaving the user (at - least me) puzzled. + holes` is in effect, only the type error is reported and the scope error + just vanishes, leaving the user (at least me) puzzled. New description: The following code snippet is well-typed, and GHC is fine with it: {{{#!hs {-# LANGUAGE TypeFamilies #-} import Control.Monad -- comment this out to cause error import Data.IORef class MonadRef m where type Ref m :: * -> * newRef :: a -> m (Ref m a) readRef :: Ref m a -> m a instance MonadRef IO where type Ref IO = IORef newRef = newIORef readRef = readIORef main :: IO () main = newRef (pure ()) >>= join . readRef }}} However, if one removes the import of `Control.Monad`, then GHC treats `join` as a typed hole due to #10569, but fails to infer its type, causing errors: {{{ test.hs:17:8-23: error: … • Couldn't match type ‘Ref m0’ with ‘IORef’ ... test.hs:17:29-32: error: … Variable not in scope: join :: m0 (f0 ()) -> IO () }}} By default the not-in-scope error is reported, but if `-fdefer-typed- holes` is in effect, only the type error is reported and the scope error just vanishes, leaving the user (at least me) puzzled. Not-in-scope errors should ''always'' be reported. -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12406#comment:1 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler