PROGRAMMING IN HASKELL Functors and monads Based on

PROGRAMMING IN HASKELL Functors and monads Based on lecture notes by Graham Hutton The book “Learn You a Haskell for Great Good” (and a few other sources) 0

Last time: Functors are a typeclass, just like Ord, Eq, Show, and all the others. This one is designed to hold things that can be mapped over; for example, lists are part of this typeclass Functor f where fmap : : (a -> b) -> f a -> f b Essentially, fmap promotes an “ordinary” function, that takes a -> b, to a function that works over values in a context. 1

Context A context just means some extra information or effects surrounding some otherwise normal value. For example: • Double indicates a floating-point number (no context) • [Double] is a floating-point number that may have many (or zero) values • Maybe Double is a floating-point number that may or may not exist • Writer String Double is a floating-point number that may have had notes (of type String) attached to it by previous computations 2

Last time: we started a tree class. (I promise we’ll get to functors again in a minute!) You can pull the code for this off my webpage, next to the lecture slides, or use yours from last time. data Tree a = Empty. Tree | Node a (Tree a) deriving (Show, Read, Eq) 3

An example run: ghci> let nums = [8, 6, 4, 1, 7, 3, 5] ghci> let nums. Tree = foldr tree. Insert Empty. Tre e nums ghci> nums. Tree Node 5 (Node 3 (Node 1 Empty. Tree) (N ode 4 Empty. Tree)) (Node 7 (Node 6 Em pty. Tree Empty. Tree) (Node 8 Empty. Tree Empty. Tre e)) 4

Back to functors: If we looked at fmap as though it were only for trees, it would look something like: (a -> b) -> Tree a -> Tree b We can certainly phrase this as a functor, also: instance Functor Tree where fmap f Empty. Tree = Empty. Tree fmap f (Node x leftsub rightsub) = Node (f x) (fmap f leftsub) (fmap f rightsub) 5

Using the tree functor: ghci> fmap (*2) Empty. Tree ghci> fmap (*4) (foldr tree. Insert Empty. Tree [5, 7, 3, 2, 1, 7]) Node 28 (Node 4 Empty. Tree (Node 8 Empty. Tree (N ode 12 Empty. Tree (Node 20 Empty. Tree )))) Empty. Tree 6

Another functor: IO actions main = do line <- get. Line let line' = reverse line put. Str. Ln $ "You said " ++ line' ++ " backwards!" put. Str. Ln $ "Yes, you really said" ++ line' ++ " backwards!" In the above code, we are getting a line as an IO action, then reversing it and printing it back out. But – an IO is a functor, which means it is designed to be mapped over using fmap! 7

Another functor: IO actions Old way: main = do line <- get. Line let line' = reverse line put. Str. Ln $ "You said " ++ line' ++ " backwards!" put. Str. Ln $ "Yes, you really said" ++ line' ++ " backwards!" Better way: use fmap! This getline has type IO String, so fmap with a string function will map the function over the result of get. Line: main = do line <- fmap reverse get. Line put. Str. Ln $ "You said " ++ line ++ " backwards!" put. Str. Ln $ "Yes, you really said" ++ line ++ " backwards!" 8

Functors and IOs In general, any time you are binding an IO action to a name, only to call functions on that name, use fmap instead! import Data. Char import Data. List main = do line <- fmap (intersperse '-'. reverse. map to. Upper) get. Line put. Str. Ln line $ runhaskell fmapping_io. hs hello there E-R-E-H-T- -O-L-L-E-H 9

How do these type? ghci> : t fmap (*2) : : (Num a, Functor f) => f a -> f a ghci> : t fmap (replicate 3) : : (Functor f) => f a -> f [a] • The expression fmap (*2) is a function that takes a functor f over numbers and returns a functor over numbers. • That functor can be a list, a Maybe , an Either String, whatever. • The expression fmap (replicate 3) will take a functor over any type and return a functor over a list of elements of that type. 10

What will one of these do? ghci> fmap (replicate 3) [1, 2, 3, 4] [[1, 1, 1], [2, 2, 2], [3, 3, 3], [4, 4, 4]] ghci> fmap (replicate 3) (Just 4) Just [4, 4, 4] ghci> fmap (replicate 3) Nothing • The type fmap (replicate 3) : : (Functor f) => f a -> f [a] means that the function will work on any functor. What exactly it will do depends on which functor we use it on. • • If we use fmap (replicate 3) on a list, the list's implementation for fmap will be chosen, which is just map. If we use it on a Maybe a, it'll apply replicate 3 to the value inside the Just, or if it's Nothing, then it stays Nothing. 11

Functor laws There are two laws any functor MUST follow if you define them: fmap id = id fmap (g. f) = fmap g. fmap f • • • If we can show that some type obeys both functor laws, we can rely on it having the same fundamental behaviors as other functors when it comes to mapping. We can know that when we use fmap on it, there won't be anything other than mapping going on behind the scenes and that it will act like a thing that can be mapped over, i. e. a functor. This leads to code that is more abstract and extensible, because we can use laws to reason about behaviors that any functor should have and make functions that operate reliably on any functor. 12

Takeaway: WHY? • • The availability of the fmap method relieves us from having to recall, read, and write a plethora of differently named mapping methods (maybe. Map, tree. Map, weird. Map, ad infinitum). As a consequence, code becomes both cleaner and easier to understand. On spotting a use of fmap, we instantly have a general idea of what is going on. Thanks to the guarantees given by the functor laws, this general idea is surprisingly precise. Using the type class system, we can write fmap-based algorithms which work out of the box with any functor - be it [], Maybe, Tree or whichever you need. Indeed, a number of useful classes in the core libraries inherit from Functor. 13

Monads A monad is defined by 3 things: • A type constructor m • A function return • An operator (>>=) – read as “bind” Originally, these were introduced for IO, but they can be much more powerful (and difficult to understand). The function and operator are methods of the Monad typeclass and have types themselves, which are required to obey 3 laws (more later): return : : a -> m a (>>=) : : m a -> (a -> m b) -> m b 14

Monads example: Maybe return : : a -> Maybe a return x = Just x (>>=) : : Maybe a -> (a -> Maybe b) -> Maybe b m >>= g = case m of Nothing -> Nothing Just x -> g x • • • Maybe is the monad, and return brings a value into it by wrapping it with Just. As for (>>=), it takes a m : : Maybe a value and a g : : a -> Maybe b function. If m is Nothing, there is nothing to do and the result is Nothing. Otherwise, in the Just x case, g is applied to x, the underlying value wrapped in Just, to give a Maybe b result. To sum it all up, if there is an underlying value of type a in m, we apply g to it, which brings the underlying value back into the Maybe monad. 15

Monads: motivation Continuing this example: Imagine a family database with two functions, which look up the parents of a person: father : : Person -> Maybe Person mother : : Person -> Maybe Person The Maybe is there in case our database is missing some information. We then might want to build other functions, to look up grandparents, etc (next slide) 16

Monads: motivation Now, let’s find grandparents: maternal. Grandfather : : Person -> Maybe Person maternal. Grandfather p = case mother p of Nothing -> Nothing Just mom -> father mom But there’s a better way! Maybe is a monad, so can do: maternal. Grandfather p = mother p >>= father 17

Monads: Even more drastic: Becomes: both. Grandfathers : : Person -> Maybe (Person, Person) both. Grandfathers p = case father p of Nothing -> Nothing Just dad -> case father dad of Nothing -> Nothing Just gf 1 -> grandfather case mother p of Nothing -> Nothing Just mom -> case father mom of Nothing -> Nothing Just gf 2 -> grandfather Just (gf 1, gf 2) -- found first -- found second both. Grandfathers p = father p >>= (dad -> father dad >>= (gf 1 -> mother p >>= -- gf 1 is only used in the final return (mom -> father mom >>= (gf 2 -> return (gf 1, gf 2) )))) 18

Monads: nuts and bolts Monad the typeclass lives in Control. Monad, and has the following methods: class Applicative m => Monad m where return : : a -> m a (>>=) : : m a -> (a -> m b) -> m b (>>) fail : : m a -> m b : : String -> m a • Aside from return and bind, there are two additional methods, (>>) and fail. • Both of them have default implementations, and so you don't need to provide them when writing an instance. 19

Monads: nuts and bolts Monad is a subclass of Applicative (more later on that one), which is a subclass of Functor. So these are also functors! Note: (>>) and fail both have default implementations, and so you don't need to provide them when writing an instance. (>>) sequences two monadic actions when the second action does not involve the result of the first, which is a common scenario for monads such as IO. Simple example: print. Something. Twice : : String -> IO () print. Something. Twice str = put. Str. Ln str >> put. Str. Ln str 20

Monads: back to that example In fact, that grandfather example can be made even better if we use the do notation with braces and semi-colons. This will look more like imperative code that we’re used to! (Like IOs. ) Here, father and mother are functions that might fail to produce results, i. e. raise an exception, and when that happens, the whole do-block will fail, i. e. terminate with an exception. both. Grandfathers p = do { dad <- father p; gf 1 <- father dad; mom <- mother p; gf 2 <- father mom; return (gf 1, gf 2); } 21

Monad laws Monads are required to obey 3 laws: m >>= return x >>= f = = m f x -- right unit -- left unit (m >>= f) >>= g = m >>= (x -> f x >>= g) -- associativity Monads originally come from a branch of mathematics called Category Theory. (Fortunately, it is entirely unnecessary to understand category theory in order to understand use monads in Haskell. ) 22

Monad laws Monads are required to obey 3 laws: m >>= return x >>= f = = m f x -- right unit -- left unit (m >>= f) >>= g = m >>= (x -> f x >>= g) -- associativity Why? ? 23

Another monad: lists Lists are another commonly used monad! Consider the following code: [(x, y) | x <- [1, 2, 3] , y <- [1, 2, 3], x /= y] This is list comprehension, like many we’ve seen so far. Output? 24

Another monad: lists This code: [(x, y) | x <- [1, 2, 3] , y <- [1, 2, 3], x /= y] Is actually equivalent to the following: do x <- [1, 2, 3] y <- [1, 2, 3] True <- return (x /= y) return (x, y) Here, each <- generates a set of values which are passed to the rest of the monadic expression. So x <- calls the rest 3 times, and the result (x, y) is returned once for each evaluation. 25

Applicative type class An applicative functor is something between a “normal” functor and a monad: it allows applying functions within the functor. The primary operation is <*>: GHCi> : t (<*>) : : Applicative f => f (a -> b) -> f a -> f b For example: instance Applicative Maybe where pure = Just (Just f) <*> (Just x) = Just (f x) _ <*> _ = Nothing 26

Applicative type class Why? Makes life easier! Imagine if you wanted to sum Just 2 and Just 3 together. Brute force approach would be to extract values from the wrapper with pattern matching – annoying, since tons of Maybe checks. What we’d like is an operator something like f (a -> b) -> f a -> f b, so we can apply functions in the context of a functor. (Go check previous slide – that’s what <*> is!) 27

Applicative in action In action, it looks something like this: Prelude> (+) <$> Just 2 <*> Just 3 Just 5 Expressions such as this one are said to be written in applicative style, which is as close as we can get to regular function application while working with a functor. The other operation, pure, just wraps something up in context. 28

Applicative laws As with all of these, there are rules for this type class. pure id <*> v = v -- Identity pure f <*> pure x = pure (f x) -- Homomorphism u <*> pure y = pure ($ y) <*> u -- Interchange pure (. ) <*> u <*> v <*> w = u <*> (v <*> w) -- Composition Again, these just keep things behaving like they should. 29

Applicative versus monad In fact, some similarities come out: pure : : Applicative f => a -> f a return : : Monad m => a -> m a The only difference is the class constraint (the m in the monad one). Really, pure and return serve the same purpose; that is, bringing values into functors. 30

Applicative functor: lists We can use <*> over lists, too: Prelude> [(2*), (5*), (9*)] <*> [1, 4, 7] [2, 8, 14, 5, 20, 35, 9, 36, 63] The only difference is the class constraint (the m in the monad one). Really, pure and return serve the same purpose; that is, bringing values into functors. 31

Recap: a sliding scale of power Functor, Applicative, and monad: ignoring pure and return, we get an increasing amount of power: fmap : : Functor f => (a -> b) -> f a -> f b (<*>) : : Applicative f => f (a -> b) -> f a -> f b (>>=) : : Monad m => m a -> (a -> m b) -> m b Rewording a bit: (<$>) : : Functor t => (a -> b) -> (t a -> t b) (<*>) : : Applicative t => t (a -> b) -> (t a -> t b) (=<<) : : Monad t => (a -> t b) -> (t a -> t b) 32

Recap: a sliding scale of power What does it mean? ? (<$>) : : Functor t => (a -> b) -> (t a -> t b) (<*>) : : Applicative t => t (a -> b) -> (t a -> t b) (=<<) : : Monad t => (a -> t b) -> (t a -> t b) Armed with these functions, =<< is the only one which allows us to create contexts (a -> t b) with our inputs. Applicative allows only the combining of contexts we're bringing along. Functor allows no effects to the context at all. 33

Using these things Even if you never turn your data types into these, you use them constantly in Haskell – everything from lists on up builds upon these operations, and give Haskell its strength. However, making even basic data types in Haskell will be much easier if you give it some thought. There are many examples, from error message passing to finite state machines to complex data structures. (I’ll post a few links on the page for more reading, but most get too long to cover in a lecture!) 34