How to flatten IO (IO ())?

Question

I'm just learning Haskell and monad transformers and I've found myself with an IO (IO ()) that I'd like to flatten into just IO (). I'm sure that I'm doing something wrong, but can't pinpoint exactly where I'm getting lost.

Here is a simplified example of what I'm trying to do. This is a convoluted way of implementing echo, but it illustrates the problem.

userInput :: Monad m => ReaderT (IO String) m (IO String)
userInput = ask

echo :: Monad m => ReaderT (IO String) m (IO ())
echo = userInput >>= \input ->  -- unwrap ReaderT to get an IO String
         input >>= (\s ->       -- unwrap IO String to get a String
           putStrLn s)          -- print out the String
         & return               -- rewrap into a ReaderT

main :: IO (IO ())              -- How to turn IO (IO ()) to IO ()?
main = runReaderT echo getLine

In my real application, I have a Spock app that makes HTTP requests to an upstream server. Spock apps use a monad transformer stack called SpockCtxT and I'd like to insert a ReaderT into the stack to abstract the HTTP request so that I can swap it out for a mock implementation in my tests.

Fundamentally, the idea is a monad transformer stack where one of the transformers gives you an IO whether it be an HTTP request or getLine. Am I thinking about this incorrectly or is there some way to do this?

Answer 1

The answer to the question as asked is join :: IO (IO ()) -> IO (). But the answer to the question I think you should have asked is liftIO :: IO () -> ReaderT (IO String) IO (). Like this:

userInput :: MonadIO m => ReaderT (IO String) m String
userInput = ask >>= liftIO -- this liftIO eliminates your need for join

echo :: MonadIO m => ReaderT (IO String) m ()
echo = userInput >>= liftIO . putStrLn -- this liftIO is just so you can use putStrLn in ReaderT

main :: IO ()
main = runReaderT echo getLine

Building monadic actions that return monadic actions, and then manually combining the inner actions, is in most cases ignoring the whole point of monad transformers. Instead of having two layers of monadic actions, you should have a single layer which has a transformer version of the outer action on top of the inner action -- that is, instead of working with ReaderT r Foo (IO a) actions, which require manual binds for both the ReaderT r Foo layer and the IO layer, you should be working with ReaderT r (FooT IO) a actions, where just one binding handles the reader, foo, and IO effects at once.