On Thu, Oct 08, 2009 at 12:54:14AM -0700, oleg@okmij.org wrote:
Actually it is possible to implement all three evaluation orders within the same final tagless framework, using the same interpretation of types and reusing most of the code save the semantics of lam. That is where the three orders differ, by their own definition.
That's really nice, Oleg, thanks! I just wanted to comment that I'd prefer to write share :: IO a -> IO (IO a) share m = mdo r <- newIORef (do x <- m writeIORef r (return x) return x) return (readIORef r >>= id) which unfortunately needs {-# LANGUAGE RecursiveDo #-} or some ugliness from mfix share :: IO a -> IO (IO a) share m = do r <- mfix $ \r -> newIORef (do x <- m writeIORef r (return x) return x) return (readIORef r >>= id) Thanks, -- Felipe.