A crash course in Haskell

What is Haskell?


A typed, lazy, purely functional programming language


Haskell = λ-calculus +

  • better syntax
  • types
  • built-in features
    • booleans, numbers, characters
    • records (tuples)
    • lists
    • recursion
    • …







Why Haskell?

Haskell programs tend to be concise and correct

QuickSort in Haskell

sort []     = []
sort (x:xs) = sort ls ++ [x] ++ sort rs
  where
    ls      = [ l | l <- xs, l <= x ]
    rs      = [ r | r <- xs, x <  r ]

Goals for this week

  1. Understand the code above
  2. Understand what typed, lazy, and purely functional means (and why it’s cool)







Haskell vs λ-calculus: similarities

(1) Programs

A program is an expression (not a sequence of statements)

It evaluates to a value (it does not perform actions)

  • λ:

    (\x -> x) apple     -- =~> apple

  • Haskell:

    (\x -> x) "apple"   -- =~> "apple"

(2) Functions

Functions are first-class values:

  • can be passed as arguments to other functions
  • can be returned as results from other functions
  • can be partially applied (arguments passed one at a time)
(\f x -> f (f x)) (\z -> z + 1) 0   -- =~> 2

But: unlike λ-calculus, not everything is a function!

(3) Top-level bindings

Like in Elsa, we can name terms to use them later

Elsa:

let T    = \x y -> x
let F    = \x y -> y

let PAIR = \x y -> \b -> ITE b x y
let FST  = \p -> p T
let SND  = \p -> p F

eval fst:
 FST (PAIR apple orange)
 =~> apple

Haskell:

haskellIsAwesome = True

pair = \x y -> \b -> if b then x else y
fst = \p -> p haskellIsAwesome
snd = \p -> p False

-- In GHCi:
> fst (pair "apple" "orange")   -- "apple"

The names are called top-level variables

Their definitions are called top-level bindings







Better Syntax: Equations and Patterns

You can define function bindings using equations:

pair x y b = if b then x else y -- same as: pair = \x y b -> ...
fst p      = p True             -- same as: fst = \p -> ...
snd p      = p False            -- same as: snd = \p -> ...







A single function binding can have multiple equations with different patterns of parameters:

pair x y True  = x  -- If 3rd arg evals to True,
                    -- use this equation;
pair x y False = y  -- Otherwise, if 3rd evals to False,
                    -- use this equation.

At run time, the first equation whose pattern matches the actual arguments is chosen

For now, a pattern is:

  • a variable (matches any value)

  • or a value (matches only that value)



Same as:

pair x y True  = x  -- If 3rd arg evals to True,
                    -- use this equation;
pair x y b     = y  -- Otherwise, use this equation.



Same as:

pair x y True  = x
pair x y _     = y  -- Wildcard pattern `_` is like a variable 
                    -- but cannot be used on the right







QUIZ

Which of the following definitions of pair is not the same as the original?

pair = \x y -> \b -> if b then x else y

A. pair x y = \b -> if b then x else y

B.

pair x _ True  = x
pair _ y _     = y

C. pair x _ True = x

D.

pair x y b     = x
pair x y False = y

E. C and D


Answer: E






Equations with guards

An equation can have multiple guards (Boolean expressions):

cmpSquare x y  |  x > y*y   =  "bigger :)"
               |  x == y*y  =  "same :|"
               |  x < y*y   =  "smaller :("

Same as:

cmpSquare x y  |  x > y*y   =  "bigger :)"
               |  x == y*y  =  "same :|"
               |  otherwise =  "smaller :("










Recusion

Recursion is built-in, so you can write:

sum n = if n == 0 
          then 0 
          else n + sum (n - 1)

or you can write:

sum 0 = 0
sum n = n + sum (n - 1)









The scope of variables

Top-level variable have global scope, so you can write:

message = if haskellIsAwesome          -- this var defined below
            then "I love CSE 130"
            else "I'm dropping CSE 130"
            
haskellIsAwesome = True



Or you can write:

-- What does f compute?
f 0 = True
f n = g (n - 1) -- mutual recursion!

g 0 = False
g n = f (n - 1) -- mutual recursion!

Answer: f is isEven, g is isOdd




Is this allowed?

haskellIsAwesome = True

haskellIsAwesome = False -- changed my mind

Answer: no, a variable can be defined once per scope; no mutation!





Local variables

You can introduce a new (local) scope using a let-expression:

sum 0 = 0
sum n = let n' = n - 1          
        in n + sum n'  -- the scope of n' is the term after in




Syntactic sugar for nested let-expressions:

sum 0 = 0
sum n = let 
          n'   = n - 1
          sum' = sum n'
        in n + sum'




If you need a variable whose scope is an equation, use the where clause instead:

cmpSquare x y  |  x > z   =  "bigger :)"
               |  x == z  =  "same :|"
               |  x < z   =  "smaller :("
  where z = y*y









Types





What would Elsa say?

let WEIRDO = ONE ZERO


Answer: Nothing. When evaluated will crunch to something nonsensical. lambda-calculus is untyped.




What would Python say?

def weirdo():
  return 0(1)


Answer: Nothing. When evaluated will cause a run-time error. Python is dynamically typed.




What would Java say?

void weirdo() {
  int zero;
  zero(1);
}


Answer: Java compiler will reject this. Java is statically typed.







In Haskell every expression either has a type or is ill-typed and rejected statically (at compile-time, before execution starts)

  • like in Java
  • unlike λ-calculus or Python
weirdo = 1 0     -- rejected by GHC









Type annotations

You can annotate your bindings with their types using ::, like so:

-- | This is a Boolean:
haskellIsAwesome :: Bool            
haskellIsAwesome = True

-- | This is a string
message :: String
message = if haskellIsAwesome
            then "I love CSE 130"
            else "I'm dropping CSE 130"
            
-- | This is a word-size integer
rating :: Int
rating = if haskellIsAwesome then 10 else 0

-- | This is an arbitrary precision integer
bigNumber :: Integer
bigNumber = factorial 100

If you omit annotations, GHC will infer them for you

  • Inspect types in GHCi using :t
  • You should annotate all top-level bindings anyway! (Why?)







Function Types

Functions have arrow types:

  • \x -> e has type A -> B
  • if e has type B assuming x has type A

For example:

> :t (\x -> if x then `a` else `b`)
(\x -> if x then `a` else `b`) :: Bool -> Char





You should annotate your function bindings:

sum :: Int -> Int
sum 0 = 0
sum n = n + sum (n - 1)

With multiple arguments:

pair :: String -> (String -> (Bool -> String))
pair x y b = if b then x else y

Same as:

pair :: String -> String -> Bool -> String
pair x y b = if b then x else y







QUIZ

With pair :: String -> String -> Bool -> String, what would GHCi say to

>:t pair "apple" "orange"

A. Syntax error

B. The term is ill-typed

C. String

D. Bool -> String

E. String -> String -> Bool -> String


Answer: D








Lists

A list is

  • either an empty list

    [] -- pronounced "nil"

  • or a head element attached to a tail list

    x:xs -- pronounced "x cons xs"



Examples:

[]                -- A list with zero elements

1:[]              -- A list with one element: 1

(:) 1 []          -- Same thing: for any infix op, 
                  -- (op) is a regular function!

1:(2:(3:(4:[])))  -- A list with four elements: 1, 2, 3, 4

1:2:3:4:[]        -- Same thing (: is right associative)

[1,2,3,4]         -- Same thing (syntactic sugar)



Terminology: constructors and values

[] and (:) are called the list constructors

We’ve seen constructors before:

  • True and False are Bool constructors
  • 0, 1, 2 are… well, it’s complicated, but you can think of them as Int constructors
  • these constructions didn’t take any parameters, so we just called them values

In general, a value is a constructor applied to other values

  • examples above are list values









The Type of a List

A list has type [A] if each one of its elements has type A

Examples:

myList :: [Int]
myList = [1,2,3,4]


myList' :: [Char]                   -- or :: String
myList' = ['h', 'e', 'l', 'l', 'o'] -- or = "hello"


-- myList'' :: Type error: elements have different types!
myList'' = [1, 'h']


myList''' :: [t] -- Generic: works for any type t!
myList''' = []









Functions on lists: range

-- | List of integers from n upto m
upto :: Int -> Int -> [Int]
upto l hi
  | l > u     = []
  | otherwise = l : (upto (l + 1) u)






There’s also syntactic sugar for this!

[1..7]    -- [1,2,3,4,5,6,7]
[1,3..7]  -- [1,3,5,7]









Functions on lists: length

-- | Length of the list
length :: ???
length xs = ???














Pattern matching on lists

-- | Length of the list
length :: [Int] -> Int
length []     = 0
length (_:xs) = 1 + length xs



A pattern is either a variable (incl. _) or a value

A pattern is

  • either a variable (incl. _)
  • or a constructor applied to other patterns



Pattern matching attempts to match values against patterns and, if desired, bind variables to successful matches.






QUIZ

What happens when we match the pattern (x:xs) against the value [1]?

A. Does not match

B. x is bound to 1, and xs is unbound

C. xs is bound to [1], and x is unbound

D. x is bound to 1, xs is bound to []

E. x is bound to 1, xs is bound to [1]


Answer: D







QUIZ

Which of the following is not a pattern?

A. (1:xs)

B. (_:_:_)

C. [x]

D. [1+2,x,y]

E. all of the above


Answer: D (1+2 is a function application, not a constructor application)






Some useful library functions

-- | Is the list empty?
null :: [t] -> Bool

-- | Head of the list
head :: [t] -> t   -- careful: partial function!

-- | Tail of the list
tail :: [t] -> [t] -- careful: partial function!

-- | Length of the list
length :: [t] -> Int

-- | Append two lists
(++) :: [t] -> [t] -> [t]

-- | Are two lists equal?
(==) :: [t] -> [t] -> Bool


You can search for library functions on Hoogle!







Pairs

myPair :: (String, Int)  -- pair of String and Int
myPair = ("apple", 3)


(,) is the pair constructor



Field access:

-- Using library functions:
whichFruit = fst myPair  -- "apple"
howMany    = snd myPair  -- 3

-- Using pattern matching:
isEmpty (x, y)   =  y == 0

-- same as:
isEmpty          = \(x, y) -> y == 0

-- same as:
isEmpty p        = let (x, y) = p in y == 0

-- Now p is the whole pair and x, y are first and second:
isEmpty p@(x, y) =  y == 0



You can use pattern matching not only in equations, but also in λ-bindings and let-bindings!






QUIZ

What happens if I call the following function

f :: String -> [(String, Int)] -> Int
f _ []   = 0
f x ((k,v) : ps)
  | x == k    = v
  | otherwise = f x ps

with input f "hi" [("hi", 5), ("apple", 10)]?

A. Syntax error

B. Type error

C. First pattern matches

D. Second pattern matches and binds

x -> "hi", k -> ("hi", 5), v -> ("apple", 10)

E. Second pattern matches and binds

x -> "hi", k -> "hi", v -> 5, ps -> [("apple", 10)]

Answer: E.



What does this function do?

Answer: E. A list of pairs represents key-value pairs in a dictionary; f performs lookup by key






Tuples

Can we implement triples like in λ-calculus?





Sure! But Haskell has native support for n-tuples:

myPair   :: (String, Int)
myPair   = ("apple", 3)

myTriple :: (Bool, Int, [Int])
myTriple = (True, 1, [1,2,3])

my4tuple :: (Float, Float, Float, Float)
my4tuple = (pi, sin pi, cos pi, sqrt 2)

...

-- And also:
myUnit   :: ()
myUnit   = ()







QUIZ

Which of the following terms is ill-typed?

You can assume that (+) :: Int -> Int -> Int.

A. (\(x,y,z) -> x + y + z) (1, 2, True)

B. (\(x,y,z) -> x + y + z) (1, 2, 3, 4)

C. (\(x,y,z) -> x + y + z) [1, 2, 3]

D. (\x y z -> x + y + z) (1, 2, 3)

E. all of the above


Answer: E. If we do not assume that + only works on integers, D is actually well-typed: it’s a function that expects two more triples and will add them together.







List comprehensions

A convenient way to construct lists from other lists:

[toUpper c | c <- s]  -- Convert string s to upper case


[(i,j) | i <- [1..3],
         j <- [1..i] ] -- Multiple generators
         
[(i,j) | i <- [0..5],
         j <- [0..5],
         i + j == 5] -- Guards







QuickSort in Haskell

sort :: [Int] -> [Int]
sort []     = []
sort (x:xs) = sort ls ++ [x] ++ sort rs
  where
    ls      = [ l | l <- xs, l <= x ]
    rs      = [ r | r <- xs, x <  r ]







What is Haskell?


A typed, lazy, purely functional programming language




Haskell is statically typed

Every expression either has a type, or is ill-typed and rejected at compile time


Why is this good?

  • catches errors early
  • types are contracts (you don’t have to handle ill-typed inputs!)
  • enables compiler optimizations







Haskell is purely functional

Functional = functions are first-class values

Pure = a program is an expression that evaluates to a value

  • no side effects!

  • unlike in Python, Java, etc:

    public int f(int x) {
  calls++;                         // side effect!
  System.out.println("calling f"); // side effect!
  launchMissile();                 // side effect!
  return calls;
}

  • in Haskell, a function of type Int -> Int computes a single integer output from a single integer input and does nothing else

Referential transparency: The same expression always evaluates to the same value

  • More precisely: In a scope where x1, ..., xn are defined, all occurrences of e with FV(e) = {x1, ..., xn} have the same value


Why is this good?

  • easier to reason about (remember x++ vs ++x in C++?)
  • enables compiler optimizations
  • especially great for parallelization (e1 + e2: we can always compute e1 and e2 in parallel!)







Haskell is lazy

An expression is evaluated only when its result is needed

Example: take 2 [1 .. (factorial 100)]


        take 2 (   upto 1 (factorial 100))
=>      take 2 (   upto 1 933262154439...)
=>      take 2 (1:(upto 2 933262154439...)) -- def upto
=> 1:  (take 1 (   upto 2 933262154439...)) -- def take 3
=> 1:  (take 1 (2:(upto 3 933262154439...)) -- def upto
=> 1:2:(take 0 (   upto 3 933262154439...)) -- def take 3
=> 1:2:[]                                   -- def take 1


Why is this good?

  • can implement cool stuff like infinite lists: [1..]

    -- first n pairs of co-primes: 
take n [(i,j) | i <- [1..],
                j <- [1..i],
                gcd i j == 1]
  • encourages simple, general solutions

  • but has its problems too :(






That’s all folks!