[Haskell-cafe] Portability of MonadError

5 Jan 2009

      Hello,

One thing that's been bothering me about MonadError monads is
the non-portability of code that uses a custom Error type.  Meaning, if I
have libraries A and B that use different error types, I won't be able to
write a function func:

func = (funcA >> funcB) `catchError` (\e -> ...)

funcA :: ErrorT MyErrorA m ()

funcB :: ErrorT MyErrorB m ()

So I'm wondering whether there's a reason not to introduce a type class
hierarchy instead of custom error types to make the code more portable.
Something like this:

--------------------
import Control.Exception(IOException)
import Control.Monad.Error

class (Eq e,Show e,Error e) => MyExc e where
    myExc1 :: e
    myExc1 = strMsg "myExc1"

-- Now we can simply extend MyExc and catch all errors defined in MyExc and
-- MyExc2 in the same error-handler (see handler2 below):

class MyExc e => MyExc2 e where
    myExc2 :: e
    myExc2 = strMsg "myExc2"

-- Uses the error class MyExc
test1:: (MonadError e m, Monad m, MonadIO m, MyExc e) => m ()
test1 = do
    liftIO $ putStrLn "############ Throwing myExc1: "
    throwError myExc1 `catchError` handler1

-- Uses the error class MyExc2 that extends MyExc
test2 :: (MonadError e m,MonadIO m, MyExc2 e) => m ()
test2 = do
    liftIO $ putStrLn "\n############ Throwing myExc2: "
    throwError myExc2 `catchError` handler2

-- Uses the error type class MyExc2 but throws an error
-- already defined in MyExc
test3 :: (MonadError e m,MonadIO m,MyExc2 e) => m ()
test3 = do
    liftIO $ putStrLn "\n############ Throwing myExc1 within context MyExc2: "
    throwError myExc1 `catchError` handler2

-- Error handler for class MyExc
handler1 :: (MonadError e m, MonadIO m,Monad m,MyExc e) => e -> m ()
handler1 e = do
        when (e == myExc1) $
            liftIO $ putStrLn $ "Caught a MyExc1 " ++ show e

-- Error handler for class MyExc2 (catches errors in MyExc1)
handler2 :: (MonadError e m, MonadIO m,Monad m,MyExc2 e) => e -> m ()
handler2 e = do
        when (e == myExc1) $  do
            liftIO $ putStrLn $ "Caught a MyExc1 " ++ show e
            throwError e
        when (e == myExc2) $
            liftIO $ putStrLn $ "Caught a MyExc2 " ++ show e

-- To run the code in the IO monad we need:
instance MyExc IOException where
    myExc1        = userError "myExc1 has occurred"
instance MyExc2 IOException where
    myExc2        = userError "myExc2 has occurred"

-- Run test1 and test2 in the IO monad
mymain :: IO ()
mymain = (test1 >> test2 >> test3)
         `catchError` (\e ->
             putStrLn $ "Something went wrong...\n" ++ show e)

-- Now let's try a custom monad:
newtype MyMonad e a = MyMonad { runMM :: ErrorT e IO a }
                     deriving(Monad,MonadError e,MonadIO)

runMyMonad :: Error e => MyMonad e a -> IO (Either e a)
runMyMonad = runErrorT . runMM

mymainT :: IO ()
mymainT = do
    res <- runMyMonad (test1 >> test2 >> test3 :: MyMonad IOException ())
    case res of
        Left e -> putStrLn $ "Something went wrong...\n" ++ show e
        _      -> return ()

----------------------

Maybe I'm missing something but is there any advantage of using
custom data types rather than the typeclass approach?

Cheers,
Peter

PS: Please be frank if I'm reinventing the wheel here... :-)

[Haskell-cafe] Portability of MonadError

Peter Robinson