Transformers for debugging

By Michal Kawalec @monad_cat

The plan

• What are monad transformers?
• How can they be used for debugging effectful functions?
• A transformers-based debugging pattern

Why transformers

• When we want capabilities of multiple monads
• But we can just do newtype MaybeIO a = MaybeIO {runMaybe :: IO (Maybe a)}, right?
• Code reuse, quadratic growth of the number of instances!

The simplest transformer

newtype IdentityT f a = IdentityT {runIdentityT :: f a}

which is, without record syntax:

newtype IdentityT f a = IdentityT (f a)

Instances

    
    newtype IdentityT m a = IdentityT {runIdentityT :: m a}

    instance (Functor m) => Functor (IdentityT m) where
        fmap f (IdentityT fa) = IdentityT (fmap f fa)

    instance (Applicative m) => Applicative (IdentityT m) where
        pure x = IdentityT (pure x)
        (IdentityT fab) <*> (IdentityT fa) = IdentityT (fab <*> fa)

    instance (Monad m) => Monad (IdentityT m) where
        return = pure
        (>>=) :: IdentityT m a ->
                (a -> IdentityT m b) ->
                IdentityT m b
        (IdentityT ma) >>= f =
            IdentityT $ ma >>= runIdentityT . f
    

Something more complex

    
    newtype MaybeT m a = MaybeT {runMaybeT :: m (Maybe a)}
    instance (Monad a) => Monad (MaybeT m) where
        return = pure
        (MaybeT ma) >>= f =
            MaybeT $ do
                v <- ma
                case v of
                    Nothing -> return Nothing
                    Just y -> runMaybeT (f y)

    

The complexestest

    
    newtype StateT s m a = StateT {runStateT :: s -> m (a, s)}
    instance (Monad m) => Monad (StateT s m) where
        return a = StateT $ \s -> return (a,s)

        (>>=) :: StateT s m a ->
                 (a -> StateT s m b) ->
                 StateT s m b
        (StateT x) >>= f = StateT $ \s -> do
            (v,s') <- x s
            runStateT (f v) s'

    

Testing in IO

• it's easy to test pure code
• what if most of our code is impure (glue type protocol library)?
• split the code into pure/impure parts and testing separately sucks

Example testable IO function

    
    getParsedChunk :: Connection ->
                      (BSC.ByteString -> Result ParseResult) ->
                      IO ParseResult
    getParsedChunk conn parser = do
      (parsed, cont) <- connectionGetChunk'' conn $ parseChunk parser

      if isJust cont
        then getParsedChunk conn $ fromJust cont
        else return . fromJust $ parsed
    

What can we do

• Fake the connection object? Not quite
• Can we overload getConnectionChunk''?

Introducing multiverse

    
    class Monad m => Universe m where
      connectionPut' :: Connection -> BSC.ByteString -> m ()
      connectionGetChunk'' :: Connection ->
        (BSC.ByteString -> (a, BSC.ByteString)) -> m a

    instance Universe IO where
      connectionPut' = connectionPut
      connectionGetChunk'' = connectionGetChunk'
    

getParsedChunk in Multiverse

    
    getParsedChunk :: (MonadIO m, Universe m) => Connection ->
                      (BSC.ByteString -> Result ParseResult) ->
                      m ParseResult
    getParsedChunk conn parser = do
      (parsed, cont) <- connectionGetChunk'' conn $ parseChunk parser

      if isJust cont
        then getParsedChunk conn $ fromJust cont
        else return . fromJust $ parsed
    

An universe for tests

    
    instance {-# OVERLAPPING #-} Universe (S.StateT FakeState IO) where
      connectionPut' = testConnectionPut
      connectionGetChunk'' = testConnectionGetChunk

    data FakeState = FS {
      bytesWritten :: TVar BS.ByteString,
      bytesToRead :: TVar BS.ByteString,
      reactToInput :: (BS.ByteString -> BS.ByteString)
    }
    

The mock

    
    testConnectionGetChunk :: Connection ->
        (BS.ByteString -> (a, BS.ByteString)) ->
        S.StateT FakeState IO a
    testConnectionGetChunk c proc = do
      st <- S.get
      toRead <- liftIO . atomically $ do
        bytes <- readTVar . bytesToRead $ st
        if (BS.length bytes) == 0 then retry else return bytes

      let (result, left) = proc toRead
      liftIO . atomically $ writeTVar (bytesToRead st) left
      return result
   

Running a test

    
    testLoginFailure :: IO ()
    testLoginFailure = do
      conn <- getConn
      let testState = defState
      atomically . writeTVar . bytesToRead $ testState "NO [ALERT] Invalid credentials (Failure)"

      (res, _) <- flip runStateT testState $ do
        login conn "a" "b"

      resultState res @?= NO
    

That's all

• The real code is at mkawalec/imap
• Follow me on Twitter @monad_cat