Monad Transformers
Table of Contents
A general application may have more than one effect to model, for example, doing IO while reading configurations (which is modeled by IO and Reader respectively). In this case, a single monad may not provide all of the capabilities. The Monad Transformers let us combine monads into a larger monadic computation.
1. Monad Transformers
A transformer takes an existing monad and adds another monad to it. Usually, they are suffixed with T, e.g.,
StateT s M a
stands for a stateful computation layered on top of an underlying monad M. Conceptually, if ordinary monad models s -> (a, s), then monad transformer models s -> m (a, s).
1.1. General lift
The general transformer operation is lift, which lifts an action from the underlying monad into the transformed monad.
lift :: Monad m => m a -> t m a
1.2. Transformer Order Matters
2. mtl Style: Monad Transformers as Typeclasses
Sometimes, to handle the order issue, it can be suffering to manually lift monads. To handle this, Haskell solves this problem by using typeclasses.
MonadReader, MonadState, MonadError, MonadIO
Instead of nesting monads, now effects become orthogonal in that these allow operations to work through transformer stacks automatically.
asks :: MonadReader r m => (r -> a) -> m a
get :: MonadState s m => m s
throwError :: MonadError e m => e -> m a
liftIO :: MonadIO m => IO a -> m a
Naive monad transformer stacks effects vertically, meaning, the order of effects matter. And potentially, you need to manually lift or adjust.