Don't be scared! The reader monad is actually not so complicated, and has real easy-to-use utility.
There are two ways of approaching a monad: we can ask
- What does the monad do? What operations is it equipped with? What is it good for?
- How is the monad implemented? From where does it arise?
From the first approach, the reader monad is some abstract type
data Reader env a
such that
-- Reader is a monad
instance Monad (Reader env)
-- and we have a function to get its environment
ask :: Reader env env
-- finally, we can run a Reader
runReader :: Reader env a -> env -> a
So how do we use this? Well, the reader monad is good for passing (implicit) configuration information through a computation.
Any time you have a "constant" in a computation that you need at various points, but really you would like to be able to perform the same computation with different values, then you should use a reader monad.
Reader monads are also used to do what the OO people call dependency injection. For example, the negamax algorithm is used frequently (in highly optimized forms) to compute the value of a position in a two player game. The algorithm itself though does not care what game you are playing, except that you need to be able to determine what the "next" positions are in the game, and you need to be able to tell if the current position is a victory position.
import Control.Monad.Reader
data GameState = NotOver | FirstPlayerWin | SecondPlayerWin | Tie
data Game position
= Game {
getNext :: position -> [position],
getState :: position -> GameState
}
getNext' :: position -> Reader (Game position) [position]
getNext' position
= do game <- ask
return $ getNext game position
getState' :: position -> Reader (Game position) GameState
getState' position
= do game <- ask
return $ getState game position
negamax :: Double -> position -> Reader (Game position) Double
negamax color position
= do state <- getState' position
case state of
FirstPlayerWin -> return color
SecondPlayerWin -> return $ negate color
Tie -> return 0
NotOver -> do possible <- getNext' position
values <- mapM ((liftM negate) . negamax (negate color)) possible
return $ maximum values
This will then work with any finite, deterministic, two player game.
This pattern is useful even for things that are not really dependency injection. Suppose you work in finance, you might design some complicated logic for pricing an asset (a derivative say), which is all well and good and you can do without any stinking monads. But then, you modify your program to deal with multiple currencies. You need to be able to convert between currencies on the fly. Your first attempt is to define a top level function
type CurrencyDict = Map CurrencyName Dollars
currencyDict :: CurrencyDict
to get spot prices. You can then call this dictionary in your code....but wait! That won't work! The currency dictionary is immutable and so has to be the same not only for the life of your program, but from the time it gets compiled! So what do you do? Well, one option would be to use the Reader monad:
computePrice :: Reader CurrencyDict Dollars
computePrice
= do currencyDict <- ask
--insert computation here
Perhaps the most classic use-case is in implementing interpreters. But, before we look at that, we need to introduce another function
local :: (env -> env) -> Reader env a -> Reader env a
Okay, so Haskell and other functional languages are based on the lambda calculus. Lambda calculus has a syntax that looks like
data Term = Apply Term Term | Lambda String Term | Var Term deriving (Show)
and we want to write an evaluator for this language. To do so, we will need to keep track of an environment, which is a list of bindings associated with terms (actually it will be closures because we want to do static scoping).
newtype Env = Env ([(String, Closure)])
type Closure = (Term, Env)
When we are done, we should get out a value (or an error):
data Value = Lam String Closure | Failure String
So, let's write the interpreter:
interp' :: Term -> Reader Env Value
--when we have a lambda term, we can just return it
interp' (Lambda nv t)
= do env <- ask
return $ Lam nv (t, env)
--when we run into a value, we look it up in the environment
interp' (Var v)
= do (Env env) <- ask
case lookup (show v) env of
-- if it is not in the environment we have a problem
Nothing -> return . Failure $ "unbound variable: " ++ (show v)
-- if it is in the environment, then we should interpret it
Just (term, env) -> local (const env) $ interp' term
--the complicated case is an application
interp' (Apply t1 t2)
= do v1 <- interp' t1
case v1 of
Failure s -> return (Failure s)
Lam nv clos -> local ((Env ls) -> Env ((nv, clos) : ls)) $ interp' t2
--I guess not that complicated!
Finally, we can use it by passing a trivial environment:
interp :: Term -> Value
interp term = runReader (interp' term) (Env [])
And that is it. A fully functional interpreter for the lambda calculus.
The other way to think about this is to ask: How is it implemented? The answer is that the reader monad is actually one of the simplest and most elegant of all monads.
newtype Reader env a = Reader {runReader :: env -> a}
Reader is just a fancy name for functions! We have already defined runReader
so what about the other parts of the API? Well, every Monad
is also a Functor
:
instance Functor (Reader env) where
fmap f (Reader g) = Reader $ f . g
Now, to get a monad:
instance Monad (Reader env) where
return x = Reader (\_ -> x)
(Reader f) >>= g = Reader $ x -> runReader (g (f x)) x
which is not so scary. ask
is really simple:
ask = Reader $ x -> x
while local
isn't so bad:
local f (Reader g) = Reader $ x -> runReader g (f x)
Okay, so the reader monad is just a function. Why have Reader at all? Good question. Actually, you don't need it!
instance Functor ((->) env) where
fmap = (.)
instance Monad ((->) env) where
return = const
f >>= g = x -> g (f x) x
These are even simpler. What's more, ask
is just id
and local
is just function composition with the order of the functions switched!