Say I have a monadic stack like this:
import Control.Monad.Trans.Reader
import Control.Monad.Trans.Except
import Control.Monad.Trans
type MyMonad = ReaderT Env (ExceptT String IO) -- Env is irrelevant
And a function (simplified, but the idea holds):
f :: Integer -> MyMonad Integer
f 42 = lift $ throwE "42 is an ILLEGAL number"
f n = return n
What I now want to do is call f from another function, but catch the thrown exception if it occurs and somehow handle it (for instance, throw another exception but with the message changed). I'm having a hard time figuring out what kind of lift operations should be done here for it to be done properly. I tried something like this:
g n = do
x <- (f n) `catchE'` (\_ -> lift $ throwE "nope, still illegal")
return x
where catchE' - lift . catchE
but it obviously won't work because catchE' takes something in the ExceptT monad, not MyMonad. Can it be done easily? Perhaps changing the structure of the monad stack could help?
You need more than lift to lift catch through a monad transformer. In fact, there are transformers with no way to lift catch at all (such as ContT). However for ReaderT there is, and the easiest way to make use of that is via Control.Monad.Error.catchError from the mtl library.
Related
Sometimes I need to use several nested MonadTrans. For example, I would put a MaybeT inside of a ExceptT, to mimick continue and break in imperative programming:
runExceptT . forM [1..] $ \ _ -> runMaybeT $ do
...
mzero -- this mimics continue
lift $ throwE "..." -- this mimics break
lift . lift $ putStrLn "Hello!"
...
However, as the above code shows, every time I need to do any IO inside of this "artificial loop", I need to put an ugly lift . lift before it. Imagine if I have even more complicated nestings, and a lot of IO operations, this quickly becomes an anonyance. How I make the code cleaner and more concise?
For the particular case of IO, you can use liftIO :: MonadIO m => IO a -> m a to lift an IO action through an arbitrarily deep monad transformer stack:
liftIO $ putStrLn "Hello"
This expression has type MonadIO m => m (). Any transformer stack that contains an instance of MonadIO somewhere inside of it should have a MonadIO instance itself, which is why this works for a stack of any depth.
Could someone give a super simple (few lines) monad transformer example, which is non-trivial (i.e. not using the Identity monad - that I understand).
For example, how would someone create a monad that does IO and can handle failure (Maybe)?
What would be the simplest example that would demonstrate this?
I have skimmed through a few monad transformer tutorials and they all seem to use State Monad or Parsers or something complicated (for a newbee). I would like to see something simpler than that. I think IO+Maybe would be simple, but I don't really know how to do that myself.
How could I use an IO+Maybe monad stack?
What would be on top? What would be on bottom? Why?
In what kind of use case would one want to use the IO+Maybe monad or the Maybe+IO monad? Would that make sense to create such a composite monad at all? If yes, when, and why?
This is available here as a .lhs file.
The MaybeT transformer will allow us to break out of a monad computation much like throwing an exception.
I'll first quickly go over some preliminaries. Skip down to Adding Maybe powers to IO for a worked example.
First some imports:
import Control.Monad
import Control.Monad.Trans
import Control.Monad.Trans.Maybe
Rules of thumb:
In a monad stack IO is always on the bottom.
Other IO-like monads will also, as a rule, always appear on the bottom, e.g. the state transformer monad ST.
MaybeT m is a new monad type which adds the power of the Maybe monad to the monad m - e.g. MaybeT IO.
We'll get into what that power is later. For now, get used to thinking of MaybeT IO as the maybe+IO monad stack.
Just like IO Int is a monad expression returning an Int, MaybeT IO Int is a MaybeT IO expression returning an Int.
Getting used to reading compound type signatures is half the battle to understanding monad transformers.
Every expression in a do block must be from the same monad.
I.e. this works because each statement is in the IO-monad:
greet :: IO () -- type:
greet = do putStr "What is your name? " -- IO ()
n <- getLine -- IO String
putStrLn $ "Hello, " ++ n -- IO ()
This will not work because putStr is not in the MaybeT IO monad:
mgreet :: MaybeT IO ()
mgreet = do putStr "What is your name? " -- IO monad - need MaybeT IO here
...
Fortunately there is a way to fix this.
To transform an IO expression into a MaybeT IO expression use liftIO.
liftIO is polymorphic, but in our case it has the type:
liftIO :: IO a -> MaybeT IO a
mgreet :: MaybeT IO () -- types:
mgreet = do liftIO $ putStr "What is your name? " -- MaybeT IO ()
n <- liftIO getLine -- MaybeT IO String
liftIO $ putStrLn $ "Hello, " ++ n -- MaybeT IO ()
Now all of the statement in mgreet are from the MaybeT IO monad.
Every monad transformer has a "run" function.
The run function "runs" the top-most layer of a monad stack returning
a value from the inside layer.
For MaybeT IO, the run function is:
runMaybeT :: MaybeT IO a -> IO (Maybe a)
Example:
ghci> :t runMaybeT mgreet
mgreet :: IO (Maybe ())
ghci> runMaybeT mgreet
What is your name? user5402
Hello, user5402
Just ()
Also try running:
runMaybeT (forever mgreet)
You'll need to use Ctrl-C to break out of the loop.
So far mgreet doesn't do anything more than what we could do in IO.
Now we'll work on an example which demonstrates the power of mixing
the Maybe monad with IO.
Adding Maybe powers to IO
We'll start with a program which asks some questions:
askfor :: String -> IO String
askfor prompt = do
putStr $ "What is your " ++ prompt ++ "? "
getLine
survey :: IO (String,String)
survey = do n <- askfor "name"
c <- askfor "favorite color"
return (n,c)
Now suppose we want to give the user the ability to end the survey
early by typing END in response to a question. We might handle it
this way:
askfor1 :: String -> IO (Maybe String)
askfor1 prompt = do
putStr $ "What is your " ++ prompt ++ " (type END to quit)? "
r <- getLine
if r == "END"
then return Nothing
else return (Just r)
survey1 :: IO (Maybe (String, String))
survey1 = do
ma <- askfor1 "name"
case ma of
Nothing -> return Nothing
Just n -> do mc <- askfor1 "favorite color"
case mc of
Nothing -> return Nothing
Just c -> return (Just (n,c))
The problem is that survey1 has the familiar staircasing issue which
doesn't scale if we add more questions.
We can use the MaybeT monad transformer to help us here.
askfor2 :: String -> MaybeT IO String
askfor2 prompt = do
liftIO $ putStr $ "What is your " ++ prompt ++ " (type END to quit)? "
r <- liftIO getLine
if r == "END"
then MaybeT (return Nothing) -- has type: MaybeT IO String
else MaybeT (return (Just r)) -- has type: MaybeT IO String
Note how all of the statemens in askfor2 have the same monad type.
We've used a new function:
MaybeT :: IO (Maybe a) -> MaybeT IO a
Here is how the types work out:
Nothing :: Maybe String
return Nothing :: IO (Maybe String)
MaybeT (return Nothing) :: MaybeT IO String
Just "foo" :: Maybe String
return (Just "foo") :: IO (Maybe String)
MaybeT (return (Just "foo")) :: MaybeT IO String
Here return is from the IO-monad.
Now we can write our survey function like this:
survey2 :: IO (Maybe (String,String))
survey2 =
runMaybeT $ do a <- askfor2 "name"
b <- askfor2 "favorite color"
return (a,b)
Try running survey2 and ending the questions early by typing END as a response to either question.
Short-cuts
I know I'll get comments from people if I don't mention the following short-cuts.
The expression:
MaybeT (return (Just r)) -- return is from the IO monad
may also be written simply as:
return r -- return is from the MaybeT IO monad
Also, another way of writing MaybeT (return Nothing) is:
mzero
Furthermore, two consecutive liftIO statements may always combined into a single liftIO, e.g.:
do liftIO $ statement1
liftIO $ statement2
is the same as:
liftIO $ do statement1
statement2
With these changes our askfor2 function may be written:
askfor2 prompt = do
r <- liftIO $ do
putStr $ "What is your " ++ prompt ++ " (type END to quit)?"
getLine
if r == "END"
then mzero -- break out of the monad
else return r -- continue, returning r
In a sense, mzero becomes a way of breaking out of the monad - like throwing an exception.
Another example
Consider this simple password asking loop:
loop1 = do putStr "Password:"
p <- getLine
if p == "SECRET"
then return ()
else loop1
This is a (tail) recursive function and works just fine.
In a conventional language we might write this as a infinite while loop with a break statement:
def loop():
while True:
p = raw_prompt("Password: ")
if p == "SECRET":
break
With MaybeT we can write the loop in the same manner as the Python code:
loop2 :: IO (Maybe ())
loop2 = runMaybeT $
forever $
do liftIO $ putStr "Password: "
p <- liftIO $ getLine
if p == "SECRET"
then mzero -- break out of the loop
else return ()
The last return () continues execution, and since we are in a forever loop, control passes back to the top of the do block. Note that the only value that loop2 can return is Nothing which corresponds to breaking out of the loop.
Depending on the situation you might find it easier to write loop2 rather than the recursive loop1.
Suppose you have to work with IO values that "may fail" in some sense, like foo :: IO (Maybe a), func1 :: a -> IO (Maybe b) and func2 :: b -> IO (Maybe c).
Manually checking for the presence of errors in a chain of binds quickly produces the dreaded "staircase of doom":
do
ma <- foo
case ma of
Nothing -> return Nothing
Just a -> do
mb <- func1 a
case mb of
Nothing -> return Nothing
Just b -> func2 b
How to "automate" this in some way? Perhaps we could devise a newtype around IO (Maybe a) with a bind function that automatically checks if the first argument is a Nothing inside IO, saving us the trouble of checking it ourselves. Something like
newtype MaybeOverIO a = MaybeOverIO { runMaybeOverIO :: IO (Maybe a) }
With the bind function:
betterBind :: MaybeOverIO a -> (a -> MaybeOverIO b) -> MaybeOverIO b
betterBind mia mf = MaybeOverIO $ do
ma <- runMaybeOverIO mia
case ma of
Nothing -> return Nothing
Just a -> runMaybeOverIO (mf a)
This works! And, looking at it more closely, we realize that we aren't using any particular functions exclusive to the IO monad. Generalizing the newtype a little, we could make this work for any underlying monad!
newtype MaybeOverM m a = MaybeOverM { runMaybeOverM :: m (Maybe a) }
And this is, in essence, how the MaybeT transformer works. I have left out a few details, like how to implement return for the transformer, and how to "lift" IO values into MaybeOverM IO values.
Notice that MaybeOverIO has kind * -> * while MaybeOverM has kind (* -> *) -> * -> * (because its first "type argument" is a monad type constructor, that itself requires a "type argument").
Sure, the MaybeT monad transformer is:
newtype MaybeT m a = MaybeT {unMaybeT :: m (Maybe a)}
We can implement its monad instance as so:
instance (Monad m) => Monad (MaybeT m) where
return a = MaybeT (return (Just a))
(MaybeT mmv) >>= f = MaybeT $ do
mv <- mmv
case mv of
Nothing -> return Nothing
Just a -> unMaybeT (f a)
This will allow us to perform IO with the option of failing gracefully in certain circumstances.
For instance, imagine we had a function like this:
getDatabaseResult :: String -> IO (Maybe String)
We can manipulate the monads independently with the result of that function, but if we compose it as so:
MaybeT . getDatabaseResult :: String -> MaybeT IO String
We can forget about that extra monadic layer, and just treat it as a normal monad.
It is quite hard to formulate good questions titles as a newbie. Please make this question search friendly =)
Trying to write my first "real" Haskell program (i.e. not only Project Euler stuff), I am trying to read and parse my configuration file with nice error messages. So far, I have this:
import Prelude hiding (readFile)
import System.FilePath (FilePath)
import System.Directory (doesFileExist)
import Data.Aeson
import Control.Monad.Except
import Data.ByteString.Lazy (ByteString, readFile)
-- Type definitions without real educational value here
loadConfiguration :: FilePath -> ExceptT String IO Configuration
loadConfiguration path = do
fileContent <- readConfigurationFile "C:\\Temp\\config.json"
configuration <- parseConfiguration fileContent
return configuration
readConfigurationFile :: FilePath -> ExceptT String IO ByteString
readConfigurationFile path = do
fileExists <- liftIO $ doesFileExist path
if fileExists then do
fileContent <- liftIO $ readFile path
return fileContent
else
throwError $ "Configuration file not found at " ++ path ++ "."
parseConfiguration :: ByteString -> ExceptT String IO Configuration
parseConfiguration raw = do
let result = eitherDecode raw :: Either String Configuration
case result of
Left message -> throwError $ "Error parsing configuration file: " ++ message
Right configuration -> return configuration
This works, but the IO monad in parseConfiguration is not necessary, and should go away. But I can't just drop it, of course, and I have not yet found a way to change parseConfiguration to something pure while keeping the prettyness of loadConfiguration.
What is the correct way to write this? If this is answered in the documentation, I am sorry, but I did not find it. I think reading the hackage documentation is a skill that grows as slowly as the rest of my Haskell skills. =)
P.S.: Comments on other style mistakes are, of course, very welcome!
If you are already using mtl, then the solution given by bheklilr in his comment is a good one. Make parseConfiguration work on any monad that implements MonadError.
If for whatever reason you are not using mtl, but only transformers, then you need'll a function with a type like Monad n => Except e a -> ExceptT e n a that "hoists" an Except into an ExceptT over some monad.
We can build this function using mapExceptT :: (m (Either e a) -> n (Either e' b)) -> ExceptT e m a -> ExceptT e' n b, a function that can change the base monad of an ExceptT transformer.
Except is really ExceptT Identity, so what we want is to unwrap the Identity and return the value in the new monad:
hoistExcept :: Monad n => Except e a -> ExceptT e n a
hoistExcept = mapExceptT (return . runIdentity)
You could also define it this way:
hoistExcept :: Monad n => Except e a -> ExceptT e n a
hoistExcept = ExceptT . return . runIdentity . runExceptT
I'm trying to use the "citation-resolve" package in a Haskell project I'm working on, but I'm having trouble getting my head around using EitherT's in real code. I get that they're monad transformers, and I think I understand what that means, however I can't seem to actually work out how to use them. The toy example that represents what I'm trying to do is as follows:
module Main where
import Text.EditDistance
import Text.CSL.Input.Identifier
import Text.CSL.Reference
import Control.Monad.Trans.Class
import Control.Monad.Trans.Either
main = do
putStrLn "Resolving definition"
let resRef = runEitherT $ resolveEither "doi:10.1145/2500365.2500595"
case resRef of
Left e -> do
putStrLn ("Got error: "++ e)
Right ref -> do
putStrLn ("Added reference to database: "++ (show ref))
Here, resolveEither has the type:
resolveEither :: (HasDatabase s,
Control.Monad.IO.Class.MonadIO m,
mtl-2.1.3.1:Control.Monad.State.Class.MonadState s m)
=> String -> EitherT String m Reference
and runEitherT $ resolveEither "ref" has the type:
runEitherT $ resolveEither "ref"
:: (HasDatabase s,
Control.Monad.IO.Class.MonadIO m,
mtl-2.1.3.1:Control.Monad.State.Class.MonadState s m)
=> m (Either String Reference)
However, this gives the following error:
Main.hs:10:34:
No instance for (Control.Monad.IO.Class.MonadIO (Either [Char]))
arising from a use of ‘resolveEither’
In the first argument of ‘runEitherT’, namely
‘(resolveEither "doi:10.1145/2500365.2500595")’
In the expression:
runEitherT (resolveEither "doi:10.1145/2500365.2500595")
In an equation for ‘resRef’:
resRef = runEitherT (resolveEither "doi:10.1145/2500365.2500595")
Which I have no idea how to resolve, or work around.
Any help would be appreciated, especially pointers to tutorials dealing with monad transformers from a usage perspective, not an implementation one.
Edit:
To reflect the comments on answers by dfeuer and Christian, I still get errors if I change main to the following:
main = do
putStrLn "Resolving definition"
resRef <- runEitherT (resolveEither "doi:10.1145/2500365.2500595")
case resRef of
Left e -> do
putStrLn ("Got error: "++ e)
Right ref -> do
putStrLn ("Added reference to database: "++ (show ref))
The error I get now is:
No instance for (MonadState s0 IO)
arising from a use of ‘resolveEither’
In the first argument of ‘runEitherT’, namely
‘(resolveEither "doi:10.1145/2500365.2500595")’
In a stmt of a 'do' block:
resRef <- runEitherT (resolveEither "doi:10.1145/2500365.2500595")
In the expression:
do { putStrLn "Resolving definition";
resRef <- runEitherT (resolveEither "doi:10.1145/2500365.2500595");
case resRef of {
Left e -> do { ... }
Right ref -> do { ... } } }
I'm editing my question as well as commenting, as nice code formatting is substantially easier here than in a comment.
I believe the problem is that you're trying to pattern match on resRef when what you probably want to do is execute it and pattern match on the result.
So you should try this:
main = do
putStrLn "Resolving definition"
resRef <- runEitherT $ resolveEither "doi:10.1145/2500365.2500595"
case resRef of
Left e -> do
You've encountered one of the shortcomings of the mtl class-based approach: intimidating type errors. I think it'll be helpful to imagine what the situation would look like with normal transformers-based monad transformers. I hope this will also help you get your feet with monad transformers in general. (It looks like you already understand most of this, by the way; I'm just spelling it out.)
Giving the types is a great way to start. Here's what you had:
resolveEither :: (HasDatabase s,
MonadIO m,
MonadState s m)
=> String -> EitherT String m Reference
There's a type hidden in the constraints, s, which came back to bite you a little later. The constraints, roughly speaking, express the following: s has a database (whatever that means in context); the monad or monad stack m has IO at its base, and somewhere in the monad stack m is a StateT s layer. The simplest monad stack m satisfying those properties would be HasDatabase s => StateT s IO. So we could write this:
resolveEither' :: HasDatabase s
=> String -> EitherT String (StateT s IO) Reference
resolveEither' = resolveEither
All we've done is specify the type of m so it's no longer a variable. We don't need to do that as long as we satisfy the class constraints.
Now it's clearer that there are two layers of monad transformers. Since our main function is in the IO monad, we want to end up with a value of type IO, which we can "run", for instance using <- in do notation. I think of it as "stripping away" layers of the monad transformer, from out to in. (This is what "using" monad transformers boils down to.)
For EitherT, there's a function runEitherT :: EitherT e m a -> m (Either e a). See how the m moves from "inside" the EitherT to "outside"? For me, that's the critical intuitive observation. Similarly for StateT, there's runStateT :: StateT s m a -> s -> m (a, s).
(Incidentally, both are defined as record accessors, which is idiomatic but causes them to show up a bit oddly in Haddock and with the "wrong" type signature; it took me a while to learn to look in the "Constructor" section on Haddocks and mentally add the EitherT e m a -> etc. to the front of the signature.)
So this adds up to a general solution, which you've basically worked out: we need an appropriate value of type s (which I'll call s), then we can use flip runStateT s . runEitherT $ resolveEither "ref" which has type IO ((Either String Reference), s). (Assuming I've kept the types straight in my head, which I probably didn't. I had forgotten flip the first time.) We can then pattern-match or use fst to get to the Either, which seems to be what you really want.
If you'd like me to explicate the errors GHC was giving you, I'd be glad. Informally, it was saying that you weren't "running" or stripping off all the monad transformers. More precisely, it was observing that IO wasn't something like StateT s IO. By using runStateT and runEitherT, you force or constrain the type such that the class constraints end up satisfied. This is kind of confusing when you get things slightly wrong.
Oh, regarding an idiomatic way to write the solution: I'm not sure that a separate retEither function would be idiomatic here, because it looks like it's meddling with global state, i.e. opening some sort of database file. It depends what the library's idiom is like.
Also, by using evalStateT, you're implicitly throwing away the state after evaluation, which may or may not be a bad idea. Does the library expect you to reuse the database connection?
Finally, you have some extra parentheses and some missing type signatures; hlint will help you with those.
Okay, so I think I've worked out a solution to my original problem, which was getting a value of the type IO (Either String Reference) from the function resolveEither (which it does for the resolveDef function it provides).
So, resolveEither returns a type of
(HasDatabase s, MonadIO m, MonadState s m) => String -> EitherT String m Reference
which we can transform to one of type
(HasDatabase s, MonadIO m, MonadState s m) => String -> m (Either String Reference)
using runEitherT . resolveEither. This was where I'd got up to when I asked the question. From there, i tried looking at the source to see how the library extracted a Reference type from the function resolveEither. The library uses the following function:
resolve :: (MonadIO m, MonadState s m, HasDatabase s) => String -> m Reference
resolve = liftM (either (const emptyReference) id) . runEitherT . resolveEither
however, we want to preserve the either, i.e. removing liftM (either (const emptyReference) id)
This however gets us back to where we started, so I looked at the source again, and worked out how this function is used. In the library, the function is used within the following, which transforms the output type of resolve from a value of type (MonadIO m, MonadState s m, HasDatabase s) => m Reference to one of type IO Reference:
resolveDef :: String -> IO Reference
resolveDef url = do
fn <- getDataFileName "default.db"
let go = withDatabaseFile fn $ resolve url
State.evalStateT go (def :: Database)
We can replace resolve in the previous with runEitherT.resolveEither to get a function that returns a IO (Either String Reference):
retEither s = do
fn <- getDataFileName "default.db"
let go = withDatabaseFile fn $ ( (runEitherT.resolveEither) s)
State.evalStateT go (Database Map.empty)
(I've replaced (def :: Database) with (Database Map.empty) as def is only defined internally in citation-resolve)
The overall solution then becomes:
module Main where
import Text.EditDistance
import Text.CSL.Input.Identifier.Internal
import Text.CSL.Input.Identifier
import Text.CSL.Reference
import Control.Monad.Trans.Either
import Control.Monad.State as State
import qualified Data.Map.Strict as Map
main = do
putStrLn "Resolving definition"
resRef <- retEither "doi:10.1145/2500365.2500595"
case resRef of
Left e -> putStrLn ("Got error: "++ e)
Right ref -> putStrLn ("Added reference to database: "++ (show ref))
retEither s = do
fn <- getDataFileName "default.db"
let go = withDatabaseFile fn $ ((runEitherT.resolveEither) s)
State.evalStateT go (Database Map.empty)
Which solves the original problem!
Any pointers on style, or ways of simplifying the whole process would however be very much appreciated.
I have the following code:
import Control.Monad
import Control.Monad.Trans
import Control.Monad.Trans.State
type T = StateT Int IO Int
someMaybe = Just 3
f :: T
f = do
x <- get
val <- lift $ do
val <- someMaybe
-- more code in Maybe monad
-- return 4
return 3
When I use do notation inside to work in Maybe monad it fails. From the error it gives it looks like type signature for this do doesn't match. However I have no idea how to fix it. I tried some lift combinations, but none of them worked and I don't want to guess anymore.
The problem is that Maybe is not part of your transformer stack. If your transformer only knows about StateT Int and IO, it does not know anything about how to lift Maybe.
You can fix this by changing your type T to something like:
type T = StateT Int (MaybeT IO) Int
(You'll need to import Control.Monad.Trans.Maybe.)
You will also need to change your inner do to work with MaybeT rather than Maybe. This means wrapping raw Maybe a values with MaybeT . return:
f :: T
f = do
x <- get
val <- lift $ do
val <- MaybeT $ return someMaybe
-- more code in Maybe monad
return 4
return 3
This is a little awkward, so you probably want to write a function like liftMaybe:
liftMaybe = MaybeT . return
If you used lift to lift IO a values in other parts of your code, this will now break because you have three levels in your transformer stack now. You will get an error that looks like this:
Couldn't match expected type `MaybeT IO t0'
with actual type `IO String'
To fix this, you should use liftIO for all your raw IO a values. This uses a typeclass to life IO actions through any number of transformer layers.
In response to your comment: if you only have a bit of code depending on Maybe, it would be easier just to put the result of the do notation into a variable and match against that:
let maybeVal = do val <- someMaybe
-- more Maybe code
return 4
case maybeVal of
Just res -> ...
Nothing -> ...
This means that the Maybe code will not be able to do an IO. You can also naturally use a function like fromMaybe instead of case.
If you want to run the code in the inner do purely in the Maybe monad, you will not have access to the StateT Int or IO monads (which might be a good thing). Doing so will return a Maybe value, which you will have to scrutinize:
import Control.Monad
import Control.Monad.Trans
import Control.Monad.Trans.State
type T = StateT Int IO Int
someMaybe = Just 3
f :: T
f = do
x <- get
-- no need to use bind
let mval = do
-- this code is purely in the Maybe monad
val <- someMaybe
-- more code in Maybe monad
return 4
-- scrutinize the resulting Maybe value now we are back in the StateT monad
case mval of
Just val -> liftIO . putStrLn $ "I got " ++ show val
Nothing -> liftIO . putStrLn $ "I got a rock"
return 3