Context
Most Haskell tutorials I know (e.g. LYAH) introduce newtypes as a cost-free idiom that allows enforcing more type safety. For instance, this code will type-check:
type Speed = Double
type Length = Double
computeTime :: Speed -> Length -> Double
computeTime v l = l / v
but this won't:
newtype Speed = Speed { getSpeed :: Double }
newtype Length = Length { getLength :: Double }
-- wrong!
computeTime :: Speed -> Length -> Double
computeTime v l = l / v
and this will:
-- right
computeTime :: Speed -> Length -> Double
computeTime (Speed v) (Length l) = l / v
In this particular example, the compiler knows that Speed is just a Double, so the pattern-matching is moot and will not generate any executable code.
Question
Are newtypes still cost-free when they appear as arguments of parametric types? For instance, consider a list of newtypes:
computeTimes :: [Speed] -> Length -> [Double]
computeTimes vs l = map (\v -> getSpeed v / l) vs
I could also pattern-match on speed in the lambda:
computeTimes' :: [Speed] -> Length -> [Double]
computeTimes' vs l = map (\(Speed v) -> v / l) vs
In either case, for some reason, I feel that real work is getting done! I start to feel even more uncomfortable when the newtype is buried within a deep tree of nested parametric datatypes, e.g. Map Speed [Set Speed]; in this situation, it may be difficult or impossible to pattern-match on the newtype, and one would have to resort to accessors like getSpeed.
TL;DR
Will the use of a newtype never ever incur a cost, even when the newtype appears as a (possibly deeply-buried) argument of another parametric type?
On their own, newtypes are cost-free. Applying their constructor, or pattern matching on them has zero cost.
When used as parameter for other types e.g. [T] the representation of [T] is precisely the same as the one for [T'] if T is a newtype for T'. So, there's no loss in performance.
However, there are two main caveats I can see.
newtypes and instances
First, newtype is frequently used to introduce new instances of type classes. Clearly, when these are user-defined, there's no guarantee that they have the same cost as the original instances. E.g., when using
newtype Op a = Op a
instance Ord a => Ord (Op a) where
compare (Op x) (Op y) = compare y x
comparing two Op Int will cost slightly more than comparing Int, since the arguments need to be swapped. (I am neglecting optimizations here, which might make this cost free when they trigger.)
newtypes used as type arguments
The second point is more subtle. Consider the following two implementations of the identity [Int] -> [Int]
id1, id2 :: [Int] -> [Int]
id1 xs = xs
id2 xs = map (\x->x) xs
The first one has constant cost. The second has a linear cost (assuming no optimization triggers). A smart programmer should prefer the first implementation, which is also simpler to write.
Suppose now we introduce newtypes on the argument type, only:
id1, id2 :: [Op Int] -> [Int]
id1 xs = xs -- error!
id2 xs = map (\(Op x)->x) xs
We can no longer use the constant cost implementation because of a type error. The linear cost implementation still works, and is the only option.
Now, this is quite bad. The input representation for [Op Int] is exactly, bit by bit, the same for [Int]. Yet, the type system forbids us to perform the identity in an efficient way!
To overcome this issue, safe coercions where introduced in Haskell.
id3 :: [Op Int] -> [Int]
id3 = coerce
The magic coerce function, under certain hypotheses, removes or inserts newtypes as needed to make type match, even inside other types, as for [Op Int] above. Further, it is a zero-cost function.
Note that coerce works only under certain conditions (the compiler checks for them). One of these is that the newtype constructor must be visible: if a module does not export Op :: a -> Op a you can not coerce Op Int to Int or vice versa. Indeed, if a module exports the type but not the constructor, it would be wrong to make the constructor accessible anyway through coerce. This makes the "smart constructors" idiom still safe: modules can still enforce complex invariants through opaque types.
It doesn't matter how deeply buried a newtype is in a stack of (fully) parametric types. At runtime, the values v :: Speed and w :: Double are completely indistinguishable – the wrapper is erased by the compiler, so even v is really just a pointer to a single 64-bit floating-point number in memory. Whether that pointer is stored in a list or tree or whatever doesn't make a difference either. getSpeed is a no-op and will not appear at runtime in any way at all.
So what do I mean by “fully parametric”? The thing is, newtypes can obviously make a difference at compile time, via the type system. In particular, they can guide instance resolution, so a newtype that invokes a different class method may certainly have worse (or, just as easily, better!) performance than the wrapped type. For example,
class Integral n => Fibonacci n where
fib :: n -> Integer
instance Fibonacci Int where
fib = (fibs !!)
where fibs = [ if i<2 then 1
else fib (i-2) + fib (i-1)
| i<-[0::Int ..] ]
this implementation is pretty slow, because it uses a lazy list (and performs lookups in it over and over again) for memoisation. On the other hand,
import qualified Data.Vector as Arr
-- | A number between 0 and 753
newtype SmallInt = SmallInt { getSmallInt :: Int }
instance Fibonacci SmallInt where
fib = (fibs Arr.!) . getSmallInt
where fibs = Arr.generate 754 $
\i -> if i<2 then 1
else fib (SmallInt $ i-2) + fib (SmallInt $ i-1)
This fib is much faster, because thanks to the input being limited to a small range, it is feasible to strictly allocate all of the results and store them in a fast O (1) lookup array, not needing the spine-laziness.
This of course applies again regardless of what structure you store the numbers in. But the different performance only comes about because different method instantiations are called – at runtime this means simply, completely different functions.
Now, a fully parametric type constructor must be able to store values of any type. In particular, it cannot impose any class restrictions on the contained data, and hence also not call any class methods. Therefore this kind of performance difference can not happen if you're just dealing with generic [a] lists or Map Int a maps. It can, however, occur when you're dealing with GADTs. In this case, even the actual memory layout might be completely differet, for instance with
{-# LANGUAGE GADTs #-}
import qualified Data.Vector as Arr
import qualified Data.Vector.Unboxed as UArr
data Array a where
BoxedArray :: Arr.Vector a -> Array a
UnboxArray :: UArr.Unbox a => UArr.Vector a -> Array a
might allow you to store Double values more efficiently than Speed values, because the former can be stored in a cache-optimised unboxed array. This is only possible because the UnboxArray constructor is not fully parametric.
In my humble opinion the answers to the famous question "What is a monad?", especially the most voted ones, try to explain what is a monad without clearly explaining why monads are really necessary. Can they be explained as the solution to a problem?
Why do we need monads?
We want to program only using functions. ("functional programming (FP)" after all).
Then, we have a first big problem. This is a program:
f(x) = 2 * x
g(x,y) = x / y
How can we say what is to be executed first? How can we form an ordered sequence of functions (i.e. a program) using no more than functions?
Solution: compose functions. If you want first g and then f, just write f(g(x,y)). This way, "the program" is a function as well: main = f(g(x,y)). OK, but ...
More problems: some functions might fail (i.e. g(2,0), divide by 0). We have no "exceptions" in FP (an exception is not a function). How do we solve it?
Solution: Let's allow functions to return two kind of things: instead of having g : Real,Real -> Real (function from two reals into a real), let's allow g : Real,Real -> Real | Nothing (function from two reals into (real or nothing)).
But functions should (to be simpler) return only one thing.
Solution: let's create a new type of data to be returned, a "boxing type" that encloses maybe a real or be simply nothing. Hence, we can have g : Real,Real -> Maybe Real. OK, but ...
What happens now to f(g(x,y))? f is not ready to consume a Maybe Real. And, we don't want to change every function we could connect with g to consume a Maybe Real.
Solution: let's have a special function to "connect"/"compose"/"link" functions. That way, we can, behind the scenes, adapt the output of one function to feed the following one.
In our case: g >>= f (connect/compose g to f). We want >>= to get g's output, inspect it and, in case it is Nothing just don't call f and return Nothing; or on the contrary, extract the boxed Real and feed f with it. (This algorithm is just the implementation of >>= for the Maybe type). Also note that >>= must be written only once per "boxing type" (different box, different adapting algorithm).
Many other problems arise which can be solved using this same pattern: 1. Use a "box" to codify/store different meanings/values, and have functions like g that return those "boxed values". 2. Have a composer/linker g >>= f to help connecting g's output to f's input, so we don't have to change any f at all.
Remarkable problems that can be solved using this technique are:
having a global state that every function in the sequence of functions ("the program") can share: solution StateMonad.
We don't like "impure functions": functions that yield different output for same input. Therefore, let's mark those functions, making them to return a tagged/boxed value: IO monad.
Total happiness!
The answer is, of course, "We don't". As with all abstractions, it isn't necessary.
Haskell does not need a monad abstraction. It isn't necessary for performing IO in a pure language. The IO type takes care of that just fine by itself. The existing monadic desugaring of do blocks could be replaced with desugaring to bindIO, returnIO, and failIO as defined in the GHC.Base module. (It's not a documented module on hackage, so I'll have to point at its source for documentation.) So no, there's no need for the monad abstraction.
So if it's not needed, why does it exist? Because it was found that many patterns of computation form monadic structures. Abstraction of a structure allows for writing code that works across all instances of that structure. To put it more concisely - code reuse.
In functional languages, the most powerful tool found for code reuse has been composition of functions. The good old (.) :: (b -> c) -> (a -> b) -> (a -> c) operator is exceedingly powerful. It makes it easy to write tiny functions and glue them together with minimal syntactic or semantic overhead.
But there are cases when the types don't work out quite right. What do you do when you have foo :: (b -> Maybe c) and bar :: (a -> Maybe b)? foo . bar doesn't typecheck, because b and Maybe b aren't the same type.
But... it's almost right. You just want a bit of leeway. You want to be able to treat Maybe b as if it were basically b. It's a poor idea to just flat-out treat them as the same type, though. That's more or less the same thing as null pointers, which Tony Hoare famously called the billion-dollar mistake. So if you can't treat them as the same type, maybe you can find a way to extend the composition mechanism (.) provides.
In that case, it's important to really examine the theory underlying (.). Fortunately, someone has already done this for us. It turns out that the combination of (.) and id form a mathematical construct known as a category. But there are other ways to form categories. A Kleisli category, for instance, allows the objects being composed to be augmented a bit. A Kleisli category for Maybe would consist of (.) :: (b -> Maybe c) -> (a -> Maybe b) -> (a -> Maybe c) and id :: a -> Maybe a. That is, the objects in the category augment the (->) with a Maybe, so (a -> b) becomes (a -> Maybe b).
And suddenly, we've extended the power of composition to things that the traditional (.) operation doesn't work on. This is a source of new abstraction power. Kleisli categories work with more types than just Maybe. They work with every type that can assemble a proper category, obeying the category laws.
Left identity: id . f = f
Right identity: f . id = f
Associativity: f . (g . h) = (f . g) . h
As long as you can prove that your type obeys those three laws, you can turn it into a Kleisli category. And what's the big deal about that? Well, it turns out that monads are exactly the same thing as Kleisli categories. Monad's return is the same as Kleisli id. Monad's (>>=) isn't identical to Kleisli (.), but it turns out to be very easy to write each in terms of the other. And the category laws are the same as the monad laws, when you translate them across the difference between (>>=) and (.).
So why go through all this bother? Why have a Monad abstraction in the language? As I alluded to above, it enables code reuse. It even enables code reuse along two different dimensions.
The first dimension of code reuse comes directly from the presence of the abstraction. You can write code that works across all instances of the abstraction. There's the entire monad-loops package consisting of loops that work with any instance of Monad.
The second dimension is indirect, but it follows from the existence of composition. When composition is easy, it's natural to write code in small, reusable chunks. This is the same way having the (.) operator for functions encourages writing small, reusable functions.
So why does the abstraction exist? Because it's proven to be a tool that enables more composition in code, resulting in creating reusable code and encouraging the creation of more reusable code. Code reuse is one of the holy grails of programming. The monad abstraction exists because it moves us a little bit towards that holy grail.
Benjamin Pierce said in TAPL
A type system can be regarded as calculating a kind of static
approximation to the run-time behaviours of the terms in a program.
That's why a language equipped with a powerful type system is strictly more expressive, than a poorly typed language. You can think about monads in the same way.
As #Carl and sigfpe point, you can equip a datatype with all operations you want without resorting to monads, typeclasses or whatever other abstract stuff. However monads allow you not only to write reusable code, but also to abstract away all redundant detailes.
As an example, let's say we want to filter a list. The simplest way is to use the filter function: filter (> 3) [1..10], which equals [4,5,6,7,8,9,10].
A slightly more complicated version of filter, that also passes an accumulator from left to right, is
swap (x, y) = (y, x)
(.*) = (.) . (.)
filterAccum :: (a -> b -> (Bool, a)) -> a -> [b] -> [b]
filterAccum f a xs = [x | (x, True) <- zip xs $ snd $ mapAccumL (swap .* f) a xs]
To get all i, such that i <= 10, sum [1..i] > 4, sum [1..i] < 25, we can write
filterAccum (\a x -> let a' = a + x in (a' > 4 && a' < 25, a')) 0 [1..10]
which equals [3,4,5,6].
Or we can redefine the nub function, that removes duplicate elements from a list, in terms of filterAccum:
nub' = filterAccum (\a x -> (x `notElem` a, x:a)) []
nub' [1,2,4,5,4,3,1,8,9,4] equals [1,2,4,5,3,8,9]. A list is passed as an accumulator here. The code works, because it's possible to leave the list monad, so the whole computation stays pure (notElem doesn't use >>= actually, but it could). However it's not possible to safely leave the IO monad (i.e. you cannot execute an IO action and return a pure value — the value always will be wrapped in the IO monad). Another example is mutable arrays: after you have leaved the ST monad, where a mutable array live, you cannot update the array in constant time anymore. So we need a monadic filtering from the Control.Monad module:
filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a]
filterM _ [] = return []
filterM p (x:xs) = do
flg <- p x
ys <- filterM p xs
return (if flg then x:ys else ys)
filterM executes a monadic action for all elements from a list, yielding elements, for which the monadic action returns True.
A filtering example with an array:
nub' xs = runST $ do
arr <- newArray (1, 9) True :: ST s (STUArray s Int Bool)
let p i = readArray arr i <* writeArray arr i False
filterM p xs
main = print $ nub' [1,2,4,5,4,3,1,8,9,4]
prints [1,2,4,5,3,8,9] as expected.
And a version with the IO monad, which asks what elements to return:
main = filterM p [1,2,4,5] >>= print where
p i = putStrLn ("return " ++ show i ++ "?") *> readLn
E.g.
return 1? -- output
True -- input
return 2?
False
return 4?
False
return 5?
True
[1,5] -- output
And as a final illustration, filterAccum can be defined in terms of filterM:
filterAccum f a xs = evalState (filterM (state . flip f) xs) a
with the StateT monad, that is used under the hood, being just an ordinary datatype.
This example illustrates, that monads not only allow you to abstract computational context and write clean reusable code (due to the composability of monads, as #Carl explains), but also to treat user-defined datatypes and built-in primitives uniformly.
I don't think IO should be seen as a particularly outstanding monad, but it's certainly one of the more astounding ones for beginners, so I'll use it for my explanation.
Naïvely building an IO system for Haskell
The simplest conceivable IO system for a purely-functional language (and in fact the one Haskell started out with) is this:
main₀ :: String -> String
main₀ _ = "Hello World"
With lazyness, that simple signature is enough to actually build interactive terminal programs – very limited, though. Most frustrating is that we can only output text. What if we added some more exciting output possibilities?
data Output = TxtOutput String
| Beep Frequency
main₁ :: String -> [Output]
main₁ _ = [ TxtOutput "Hello World"
-- , Beep 440 -- for debugging
]
cute, but of course a much more realistic “alterative output” would be writing to a file. But then you'd also want some way to read from files. Any chance?
Well, when we take our main₁ program and simply pipe a file to the process (using operating system facilities), we have essentially implemented file-reading. If we could trigger that file-reading from within the Haskell language...
readFile :: Filepath -> (String -> [Output]) -> [Output]
This would use an “interactive program” String->[Output], feed it a string obtained from a file, and yield a non-interactive program that simply executes the given one.
There's one problem here: we don't really have a notion of when the file is read. The [Output] list sure gives a nice order to the outputs, but we don't get an order for when the inputs will be done.
Solution: make input-events also items in the list of things to do.
data IO₀ = TxtOut String
| TxtIn (String -> [Output])
| FileWrite FilePath String
| FileRead FilePath (String -> [Output])
| Beep Double
main₂ :: String -> [IO₀]
main₂ _ = [ FileRead "/dev/null" $ \_ ->
[TxtOutput "Hello World"]
]
Ok, now you may spot an imbalance: you can read a file and make output dependent on it, but you can't use the file contents to decide to e.g. also read another file. Obvious solution: make the result of the input-events also something of type IO, not just Output. That sure includes simple text output, but also allows reading additional files etc..
data IO₁ = TxtOut String
| TxtIn (String -> [IO₁])
| FileWrite FilePath String
| FileRead FilePath (String -> [IO₁])
| Beep Double
main₃ :: String -> [IO₁]
main₃ _ = [ TxtIn $ \_ ->
[TxtOut "Hello World"]
]
That would now actually allow you to express any file operation you might want in a program (though perhaps not with good performance), but it's somewhat overcomplicated:
main₃ yields a whole list of actions. Why don't we simply use the signature :: IO₁, which has this as a special case?
The lists don't really give a reliable overview of program flow anymore: most subsequent computations will only be “announced” as the result of some input operation. So we might as well ditch the list structure, and simply cons a “and then do” to each output operation.
data IO₂ = TxtOut String IO₂
| TxtIn (String -> IO₂)
| Terminate
main₄ :: IO₂
main₄ = TxtIn $ \_ ->
TxtOut "Hello World"
Terminate
Not too bad!
So what has all of this to do with monads?
In practice, you wouldn't want to use plain constructors to define all your programs. There would need to be a good couple of such fundamental constructors, yet for most higher-level stuff we would like to write a function with some nice high-level signature. It turns out most of these would look quite similar: accept some kind of meaningfully-typed value, and yield an IO action as the result.
getTime :: (UTCTime -> IO₂) -> IO₂
randomRIO :: Random r => (r,r) -> (r -> IO₂) -> IO₂
findFile :: RegEx -> (Maybe FilePath -> IO₂) -> IO₂
There's evidently a pattern here, and we'd better write it as
type IO₃ a = (a -> IO₂) -> IO₂ -- If this reminds you of continuation-passing
-- style, you're right.
getTime :: IO₃ UTCTime
randomRIO :: Random r => (r,r) -> IO₃ r
findFile :: RegEx -> IO₃ (Maybe FilePath)
Now that starts to look familiar, but we're still only dealing with thinly-disguised plain functions under the hood, and that's risky: each “value-action” has the responsibility of actually passing on the resulting action of any contained function (else the control flow of the entire program is easily disrupted by one ill-behaved action in the middle). We'd better make that requirement explicit. Well, it turns out those are the monad laws, though I'm not sure we can really formulate them without the standard bind/join operators.
At any rate, we've now reached a formulation of IO that has a proper monad instance:
data IO₄ a = TxtOut String (IO₄ a)
| TxtIn (String -> IO₄ a)
| TerminateWith a
txtOut :: String -> IO₄ ()
txtOut s = TxtOut s $ TerminateWith ()
txtIn :: IO₄ String
txtIn = TxtIn $ TerminateWith
instance Functor IO₄ where
fmap f (TerminateWith a) = TerminateWith $ f a
fmap f (TxtIn g) = TxtIn $ fmap f . g
fmap f (TxtOut s c) = TxtOut s $ fmap f c
instance Applicative IO₄ where
pure = TerminateWith
(<*>) = ap
instance Monad IO₄ where
TerminateWith x >>= f = f x
TxtOut s c >>= f = TxtOut s $ c >>= f
TxtIn g >>= f = TxtIn $ (>>=f) . g
Obviously this is not an efficient implementation of IO, but it's in principle usable.
Monads serve basically to compose functions together in a chain. Period.
Now the way they compose differs across the existing monads, thus resulting in different behaviors (e.g., to simulate mutable state in the state monad).
The confusion about monads is that being so general, i.e., a mechanism to compose functions, they can be used for many things, thus leading people to believe that monads are about state, about IO, etc, when they are only about "composing functions".
Now, one interesting thing about monads, is that the result of the composition is always of type "M a", that is, a value inside an envelope tagged with "M". This feature happens to be really nice to implement, for example, a clear separation between pure from impure code: declare all impure actions as functions of type "IO a" and provide no function, when defining the IO monad, to take out the "a" value from inside the "IO a". The result is that no function can be pure and at the same time take out a value from an "IO a", because there is no way to take such value while staying pure (the function must be inside the "IO" monad to use such value). (NOTE: well, nothing is perfect, so the "IO straitjacket" can be broken using "unsafePerformIO : IO a -> a" thus polluting what was supposed to be a pure function, but this should be used very sparingly and when you really know to be not introducing any impure code with side-effects.
Monads are just a convenient framework for solving a class of recurring problems. First, monads must be functors (i.e. must support mapping without looking at the elements (or their type)), they must also bring a binding (or chaining) operation and a way to create a monadic value from an element type (return). Finally, bind and return must satisfy two equations (left and right identities), also called the monad laws. (Alternatively one could define monads to have a flattening operation instead of binding.)
The list monad is commonly used to deal with non-determinism. The bind operation selects one element of the list (intuitively all of them in parallel worlds), lets the programmer to do some computation with them, and then combines the results in all worlds to single list (by concatenating, or flattening, a nested list). Here is how one would define a permutation function in the monadic framework of Haskell:
perm [e] = [[e]]
perm l = do (leader, index) <- zip l [0 :: Int ..]
let shortened = take index l ++ drop (index + 1) l
trailer <- perm shortened
return (leader : trailer)
Here is an example repl session:
*Main> perm "a"
["a"]
*Main> perm "ab"
["ab","ba"]
*Main> perm ""
[]
*Main> perm "abc"
["abc","acb","bac","bca","cab","cba"]
It should be noted that the list monad is in no way a side effecting computation. A mathematical structure being a monad (i.e. conforming to the above mentioned interfaces and laws) does not imply side effects, though side-effecting phenomena often nicely fit into the monadic framework.
You need monads if you have a type constructor and functions that returns values of that type family. Eventually, you would like to combine these kind of functions together. These are the three key elements to answer why.
Let me elaborate. You have Int, String and Real and functions of type Int -> String, String -> Real and so on. You can combine these functions easily, ending with Int -> Real. Life is good.
Then, one day, you need to create a new family of types. It could be because you need to consider the possibility of returning no value (Maybe), returning an error (Either), multiple results (List) and so on.
Notice that Maybe is a type constructor. It takes a type, like Int and returns a new type Maybe Int. First thing to remember, no type constructor, no monad.
Of course, you want to use your type constructor in your code, and soon you end with functions like Int -> Maybe String and String -> Maybe Float. Now, you can't easily combine your functions. Life is not good anymore.
And here's when monads come to the rescue. They allow you to combine that kind of functions again. You just need to change the composition . for >==.
Why do we need monadic types?
Since it was the quandary of I/O and its observable effects in nonstrict languages like Haskell that brought the monadic interface to such prominence:
[...] monads are used to address the more general problem of computations (involving state, input/output, backtracking, ...) returning values: they do not solve any input/output-problems directly but rather provide an elegant and flexible abstraction of many solutions to related problems. [...] For instance, no less than three different input/output-schemes are used to solve these basic problems in Imperative functional programming, the paper which originally proposed `a new model, based on monads, for performing input/output in a non-strict, purely functional language'. [...]
[Such] input/output-schemes merely provide frameworks in which side-effecting operations can safely be used with a guaranteed order of execution and without affecting the properties of the purely functional parts of the language.
Claus Reinke (pages 96-97 of 210).
(emphasis by me.)
[...] When we write effectful code – monads or no monads – we have to constantly keep in mind the context of expressions we pass around.
The fact that monadic code ‘desugars’ (is implementable in terms of) side-effect-free code is irrelevant. When we use monadic notation, we program within that notation – without considering what this notation desugars into. Thinking of the desugared code breaks the monadic abstraction. A side-effect-free, applicative code is normally compiled to (that is, desugars into) C or machine code. If the desugaring argument has any force, it may be applied just as well to the applicative code, leading to the conclusion that it all boils down to the machine code and hence all programming is imperative.
[...] From the personal experience, I have noticed that the mistakes I make when writing monadic code are exactly the mistakes I made when programming in C. Actually, monadic mistakes tend to be worse, because monadic notation (compared to that of a typical imperative language) is ungainly and obscuring.
Oleg Kiselyov (page 21 of 26).
The most difficult construct for students to understand is the monad. I introduce IO without mentioning monads.
Olaf Chitil.
More generally:
Still, today, over 25 years after the introduction of the concept of monads to the world of functional programming, beginning functional programmers struggle to grasp the concept of monads. This struggle is exemplified by the numerous blog posts about the effort of trying to learn about monads. From our own experience we notice that even at university level, bachelor level students often struggle to comprehend monads and consistently score poorly on monad-related exam questions.
Considering that the concept of monads is not likely to disappear from the functional programming landscape any time soon, it is vital that we, as the functional programming community, somehow overcome the problems novices encounter when first studying monads.
Tim Steenvoorden, Jurriën Stutterheim, Erik Barendsen and Rinus Plasmeijer.
If only there was another way to specify "a guaranteed order of execution" in Haskell, while keeping the ability to separate regular Haskell definitions from those involved in I/O (and its observable effects) - translating this variation of Philip Wadler's echo:
val echoML : unit -> unit
fun echoML () = let val c = getcML () in
if c = #"\n" then
()
else
let val _ = putcML c in
echoML ()
end
fun putcML c = TextIO.output1(TextIO.stdOut,c);
fun getcML () = valOf(TextIO.input1(TextIO.stdIn));
...could then be as simple as:
echo :: OI -> ()
echo u = let !(u1:u2:u3:_) = partsOI u in
let !c = getChar u1 in
if c == '\n' then
()
else
let !_ = putChar c u2 in
echo u3
where:
data OI -- abstract
foreign import ccall "primPartOI" partOI :: OI -> (OI, OI)
⋮
foreign import ccall "primGetCharOI" getChar :: OI -> Char
foreign import ccall "primPutCharOI" putChar :: Char -> OI -> ()
⋮
and:
partsOI :: OI -> [OI]
partsOI u = let !(u1, u2) = partOI u in u1 : partsOI u2
How would this work? At run-time, Main.main receives an initial OI pseudo-data value as an argument:
module Main(main) where
main :: OI -> ()
⋮
...from which other OI values are produced, using partOI or partsOI. All you have to do is ensure each new OI value is used at most once, in each call to an OI-based definition, foreign or otherwise. In return, you get back a plain ordinary result - it isn't e.g. paired with some odd abstract state, or requires the use of a callback continuation, etc.
Using OI, instead of the unit type () like Standard ML does, means we can avoid always having to use the monadic interface:
Once you're in the IO monad, you're stuck there forever, and are reduced to Algol-style imperative programming.
Robert Harper.
But if you really do need it:
type IO a = OI -> a
unitIO :: a -> IO a
unitIO x = \ u -> let !_ = partOI u in x
bindIO :: IO a -> (a -> IO b) -> IO b
bindIO m k = \ u -> let !(u1, u2) = partOI u in
let !x = m u1 in
let !y = k x u2 in
y
⋮
So, monadic types aren't always needed - there are other interfaces out there:
LML had a fully fledged implementation of oracles running of a multi-processor (a Sequent Symmetry) back in ca 1989. The description in the Fudgets thesis refers to this implementation. It was fairly pleasant to work with and quite practical.
[...]
These days everything is done with monads so other solutions are sometimes forgotten.
Lennart Augustsson (2006).
Wait a moment: since it so closely resembles Standard ML's direct use of effects, is this approach and its use of pseudo-data referentially transparent?
Absolutely - just find a suitable definition of "referential transparency"; there's plenty to choose from...
In Haskell, there is a function called unsafeCoerce, that turns anything into any other type of thing. What exactly is this used for? Like, why we would you want to transform things into each other in such an "unsafe" way?
Provide an example of a way that unsafeCoerce is actually used. A link to Hackage would help. Example code in someones question would not.
unsafeCoerce lets you convince the type system of whatever property you like. It's thus only "safe" exactly when you can be completely certain that the property you're declaring is true. So, for instance:
unsafeCoerce True :: Int
is a violation and can lead to wonky, bad runtime behavior.
unsafeCoerce (3 :: Int) :: Int
is (obviously) fine and will not lead to runtime misbehavior.
So what's a non-trivial use of unsafeCoerce? Let's say we've got an typeclass-bound existential type
module MyClass ( SomethingMyClass (..), intSomething ) where
class MyClass x where {}
instance MyClass Int where {}
data SomethingMyClass = forall a. MyClass a => SomethingMyClass a
Let's also say, as noted here, that the typeclass MyClass is not exported and thus nobody else can ever create instances of it. Indeed, Int is the only thing that instantiates it and the only thing that ever will.
Now when we pattern match to destruct a value of SomethingMyClass we'll be able to pull a "something" out from inside
foo :: SomethingMyClass -> ...
foo (SomethingMyClass a) =
-- here we have a value `a` with type `exists a . MyClass a => a`
--
-- this is totally useless since `MyClass` doesn't even have any
-- methods for us to use!
...
Now, at this point, as the comment suggests, the value we've pulled out has no type information—it's been "forgotten" by the existential context. It could be absolutely anything which instantiates MyClass.
Of course, in this very particular situation we know that the only thing implementing MyClass is Int. So our value a must actually have type Int. We could never convince the typechecker that this is true, but due to an outside proof we know that it is.
Therefore, we can (very carefully)
intSomething :: SomethingMyClass -> Int
intSomething (SomethingMyClass a) = unsafeCoerce a -- shudder!
Now, hopefully I've suggested that this is a terrible, dangerous idea, but it also may give a taste of what kind of information we can take advantage of in order to know things that the typechecker cannot.
In non-pathological situations, this is rare. Even rarer is a situation where using something we know and the typechecker doesn't isn't itself pathological. In the above example, we must be completely certain that nobody ever extends our MyClass module to instantiate more types to MyClass otherwise our use of unsafeCoerce becomes instantly unsafe.
> instance MyClass Bool where {}
> intSomething (SomethingMyClass True)
6917529027658597398
Looks like our compiler internals are leaking!
A more common example where this sort of behavior might be valuable is when using newtype wrappers. It's a fairly common idea that we might wrap a type in a newtype wrapper in order to specialize its instance definitions.
For example, Int does not have a Monoid definition because there are two natural monoids over Ints: sums and products. Instead, we use newtype wrappers to be more explicit.
newtype Sum a = Sum { getSum :: a }
instance Num a => Monoid (Sum a) where
mempty = Sum 0
mappend (Sum a) (Sum b) = Sum (a+b)
Now, normally the compiler is pretty smart and recognizes that it can eliminate all of those Sum constructors in order to produce more efficient code. Sadly, there are times when it cannot, especially in highly polymorphic situations.
If you (a) know that some type a is actually just a newtype-wrapped b and (b) know that the compiler is incapable of deducing this itself, then you might want to do
unsafeCoerce (x :: a) :: b
for a slight efficiency gain. This, for instance, occurs frequently in lens and is expressed in the Data.Profunctor.Unsafe module of profunctors, a dependency of lens.
But let me again suggest that you really need to know what's going on before using unsafeCoerce like this is anything but highly unsafe.
One final thing to compare is the "typesafe cast" available in Data.Typeable. This function looks a bit like unsafeCoerce, but with much more ceremony.
unsafeCoerce :: a -> b
cast :: (Typeable a, Typeable b) => a -> Maybe b
Which, you might think of as being implemented using unsafeCoerce and a function typeOf :: Typeable a => a -> TypeRep where TypeRep are unforgeable, runtime tokens which reflect the type of a value. Then we have
cast :: (Typeable a, Typeable b) => a -> Maybe b
cast a = if (typeOf a == typeOf b) then Just b else Nothing
where b = unsafeCoerce a
Thus, cast is able to ensure that the types of a and b really are the same at runtime, and it can decide to return Nothing if they are not. As an example:
{-# LANGUAGE DeriveDataTypeable #-}
{-# LANGUAGE ExistentialQuantification #-}
data A = A deriving (Show, Typeable)
data B = B deriving (Show, Typeable)
data Forget = forall a . Typeable a => Forget a
getAnA :: Forget -> Maybe A
getAnA (Forget something) = cast something
which we can run as follows
> getAnA (Forget A)
Just A
> getAnA (Forget B)
Nothing
So if we compare this usage of cast with unsafeCoerce we see that it can achieve some of the same functionality. In particular, it allows us to rediscover information that may have been forgotten by ExistentialQuantification. However, cast manually checks the types at runtime to ensure that they are truly the same and thus cannot be used unsafely. To do this, it demands that both the source and target types allow for runtime reflection of their types via the Typeable class.
The only time I ever felt compelled to use unsafeCoerce was on finite natural numbers.
{-# LANGUAGE DataKinds, GADTs, TypeFamilies, StandaloneDeriving #-}
data Nat = Z | S Nat deriving (Eq, Show)
data Fin (n :: Nat) :: * where
FZ :: Fin (S n)
FS :: Fin n -> Fin (S n)
deriving instance Show (Fin n)
Fin n is a singly linked data structure that is statically ensured to be smaller than the n type level natural number by which it is parametrized.
-- OK, 1 < 2
validFin :: Fin (S (S Z))
validFin = FS FZ
-- type error, 2 < 2 is false
invalidFin :: Fin (S (S Z))
invalidFin = FS (FS FZ)
Fin can be used to safely index into various data structures. It's pretty standard in dependently typed languages, though not in Haskell.
Sometimes we want to convert a value of Fin n to Fin m where m is greater than n.
relaxFin :: Fin n -> Fin (S n)
relaxFin FZ = FZ
relaxFin (FS n) = FS (relaxFin n)
relaxFin is a no-op by definition, but traversing the value is still required for the types to check out. So we might just use unsafeCoerce instead of relaxFin. More pronounced gains in speed can result from coercing larger data structures that contain Fin-s (for example, you could have lambda terms with Fin-s as bound variables).
This is an admittedly exotic example, but I find it interesting in the sense that it's pretty safe: I can't really think of ways for external libraries or safe user code to mess this up. I might be wrong though and I'd be eager to hear about potential safety issues.
There is no use of unsafeCoerce I can really recommend, but I can see that in some cases such a thing might be useful.
The first use that springs to mind is the implementation of the Typeable-related routines. In particular cast :: (Typeable a, Typeable b) => a -> Maybe b achieves a type-safe behaviour, so it is safe to use, yet it has to play dirty tricks in its implementation.
Maybe unsafeCoerce can find some use when importing FFI subroutines to force types to match. After all, FFI already allows to import impure C functions as pure ones, so it is intrinsecally usafe. Note that "unsafe" does not mean impossible to use, but just "putting the burden of proof on the programmer".
Finally, pretend that sortBy did not exist. Consider then this example:
-- Like Int, but using the opposite ordering
newtype Rev = Rev { unRev :: Int }
instance Ord Rev where compare (Rev x) (Rev y) = compare y x
sortDescending :: [Int] -> [Int]
sortDescending = map unRev . sort . map Rev
The code above works, but feels silly IMHO. We perform two maps using functions such as Rev,unRev which we know to be no-ops at runtime. So we just scan the list twice for no reason, but that of convincing the compiler to use the right Ord instance.
The performance impact of these maps should be small since we also sort the list. Yet it is tempting to rewrite map Rev as unsafeCoerce :: [Int]->[Rev] and save some time.
Note that having a coercing function
castNewtype :: IsNewtype t1 t2 => f t2 -> f t1
where the constraint means that t1 is a newtype for t2 would help, but it would be quite dangerous. Consider
castNewtype :: Data.Set Int -> Data.Set Rev
The above would cause the data structure invariant to break, since we are changing the ordering underneath! Since Data.Set is implemented as a binary search tree, it would cause quite a large damage.
Is it possible to implement a quicksort in Haskell (with RANDOM-PIVOT) that still has a simple Ord a => [a]->[a] signature?
I'm starting to understand Monads, and, for now, I'm kind of interpreting monads as somethink like a 'command pattern', which works great for IO.
So, I understand that a function that returns a random number should actually return a monadic value like IO, because, otherwise, it would break referential transparency. I also understand that there should be no way to 'extract' the random integer from the returned monadic value, because, otherwise, it would, again, break referential transparency.
But yet, I still think that it should be possible to implement a 'pure' [a]->[a] quicksort function, even if it uses random pivot, because, it IS referential transparent. From my point of view, the random pivot is just a implementation detail, and shouldn't change the function's signature
OBS: I'm not actually interested in the specific quicksort problem (so, I don't want to sound rude but I'm not looking for "use mergesort" or "random pivot doesn't increase performance in practice" kind of answers) I'm actually interested in how to implement a 'pure' function that uses 'impure' functions inside it, in cases like quicksort, where I can assure that the function actually is a pure one.
Quicksort is just a good example.
You are making a false assumption that picking the pivot point is just an implementation detail. Consider a partial ordering on a set. Like a quicksort on cards where
card a < card b if the face value is less but if you were to evaluate booleans:
4 spades < 4 hearts (false)
4 hearts < 4 spades (false)
4 hearts = 4 spades (false)
In that case the choice of pivots would determine the final ordering of the cards. In precisely the same way
for a function like
a = get random integer
b = a + 3
print b
is determined by a. If you are randomly choosing something then your computation is or could be non deterministic.
OK, check this out.
Select portions copied form the hashable package, and voodoo magic language pragmas
{-# LANGUAGE FlexibleInstances, UndecidableInstances, NoMonomorphismRestriction, OverlappingInstances #-}
import System.Random (mkStdGen, next, split)
import Data.List (foldl')
import Data.Bits (shiftL, xor)
class Hashable a where
hash :: a -> Int
instance (Integral a) => Hashable a where
hash = fromIntegral
instance Hashable Char where
hash = fromEnum
instance (Hashable a) => Hashable [a] where
hash = foldl' combine 0 . map hash
-- ask the authors of the hashable package about this if interested
combine h1 h2 = (h1 + h1 `shiftL` 5) `xor` h2
OK, so now we can take a list of anything Hashable and turn it into an Int. I've provided Char and Integral a instances here, more and better instances are in the hashable packge, which also allows salting and stuff.
This is all just so we can make a number generator.
genFromHashable = mkStdGen . hash
So now the fun part. Let's write a function that takes a random number generator, a comparator function, and a list. Then we'll sort the list by consulting the generator to select a pivot, and the comparator to partition the list.
qSortByGen _ _ [] = []
qSortByGen g f xs = qSortByGen g'' f l ++ mid ++ qSortByGen g''' f r
where (l, mid, r) = partition (`f` pivot) xs
pivot = xs !! (pivotLoc `mod` length xs)
(pivotLoc, g') = next g
(g'', g''') = split g'
partition f = foldl' step ([],[],[])
where step (l,mid,r) x = case f x of
LT -> (x:l,mid,r)
EQ -> (l,x:mid,r)
GT -> (l,mid,x:r)
Library functions: next grabs an Int from the generator, and produces a new generator. split forks the generator into two distinct generators.
My functions: partition uses f :: a -> Ordering to partition the list into three lists. If you know folds, it should be quite clear. (Note that it does not preserve the initial ordering of the elements in the sublists; it reverses them. Using a foldr could remedy this were it an issue.) qSortByGen works just like I said before: consult the generator for the pivot, partition the list, fork the generator for use in the two recursive calls, recursively sort the left and right sides, and concatenate it all together.
Convenience functions are easy to compose from here
qSortBy f xs = qSortByGen (genFromHashable xs) f xs
qSort = qSortBy compare
Notice the final function's signature.
ghci> :t qSort
qSort :: (Ord a, Hashable a) => [a] -> [a]
The type inside the list must implement both Hashable and Ord. There's the "pure" function you were asking for, with one logical added requirement. The more general functions are less restrictive in their requirements.
ghci> :t qSortBy
qSortBy :: (Hashable a) => (a -> a -> Ordering) -> [a] -> [a]
ghci> :t qSortByGen
qSortByGen
:: (System.Random.RandomGen t) =>
t -> (a -> a -> Ordering) -> [a] -> [a]
Final notes
qSort will behave exactly the same way for all inputs. The "random" pivot selection is. in fact, deterministic. But it is obscured by hashing the list and then seeding a random number generator, making it "random" enough for me. ;)
qSort also only works for lists with length less than maxBound :: Int, which ghci tells me is 9,223,372,036,854,775,807. I thought there would be an issue with negative indexes, but in my ad-hoc testing I haven't run into it yet.
Or, you can just live with the IO monad for "truer" randomness.
qSortIO xs = do g <- getStdGen -- add getStdGen to your imports
return $ qSortByGen g compare xs
ghci> :t qSortIO
qSortIO :: (Ord a) => [a] -> IO [a]
ghci> qSortIO "Hello world"
" Hdellloorw"
ghci> qSort "Hello world"
" Hdellloorw"
In such cases, where you know that the function is referentially transparent, but you can't proof it to the compiler, you may use the function unsafePerformIO :: IO a -> a from the module Data.Unsafe.
For instance, you may use unsafePerformIO to get an initial random state and then do anything using just this state.
But please notice: Don't use it if it's not really needed. And even then, think twice about it. unsafePerformIO is somewhat the root of all evil, since it's consequences can be dramatical - anything is possible from coercing different types to crashing the RTS using this function.
Haskell provides the ST monad to perform non-referentially-transparent actions with a referentially transparent result.
Note that it doesn't enforce referential transparency; it just insures that potentially non-referentially-transparent temporary state can't leak out. Nothing can prevent you from returning manipulated pure input data that was rearranged in a non-reproducible way. Best is to implement the same thing in both ST and pure ways and use QuickCheck to compare them on random inputs.