Delimiting the IO monad

问题:

It's nice to know (in Safe Haskell, at least) from the signature whether or not something performs IO actions, but IO encompasses a lot of different things - putStr, database access, removing and writing to files, IORefs, etc.

If I'm using the type signatures as a security measure when running arbitrary code, it might be the case that I'm willing to accept some IO actions - putStr and the ilk, for instance - but not others.

Is there a way to define a restricted version of the IO monad, with only a subset of the normal IO actions? If so, an example (with putStr, for instance) would be very welcome!

回答1:

As a follow up to my comment, you can implement it yourself with something like

class Monad io => Stdout io where
    putStr_ :: String -> io ()
    putStrLn_ :: String -> io ()
    print_ :: Show a => a -> io ()
    -- etc

instance Stdout IO where
    putStr_ = putStr
    putStrLn_ putStrLn
    print_ = print

myFunc :: Stdout io => io ()
myFunc = do
    val <- someAction
    print_ val
    let newVal = doSomething val
    print_ newVal

main :: IO ()
main = myFunc

This will have absolutely no runtime overhead, since GHC will optimize away those typeclasses to use only the IO monad, it's extensible, easy to write, and can be combined with monad transformes and the MonadIO class quite easily. If you have multiple class, such as a Stdin class with getLine_, getChar_, etc defined, you can even combine these typeclasses with

class (Stdout io, Stdin io) => StdOutIn io where

myFunc :: StdOutIn io => io ()
myFunc = do
    val <- getLine_
    putStrLn_ $ "Echo: " ++ val

main :: IO ()
main = myFunc

回答2:

Just define a newtype around IO a with a Monad instance, define wrapped versions of your pre-approved functions, and don't export the constructor, so that only the functions you wrapped can be used in the monad.