This course page was updated until March 2022 when I left Durham University. For future updates, please visit the new version of the course pages.

Higher order functions #

foldl and foldr #

Many functions that apply to lists can be written, as we saw in lectures, in the form:

f [] = v
f (x:xs) = x `op` f xs

which turns the empty list into some initial value (here v) and all other lists to the result of apply a specified binary operator op to the head of the list and the result of recursion on the tail.

For example

sum :: Num a => [a] -> a
sum [] = 0
sum (x:xs) = x + sum xs

or :: [Bool] -> Bool
or [] = False
or (x:xs) = x || or xs

We saw that this linear recursive pattern was captured in the higher order function foldr (“fold right”). It is so named because untangling the recursion, we see that the operator is applied in a right-associative manner.

foldr :: (a -> b -> b) -> b -> [a] -> b

With which we have

sum :: Num a => [a] -> a
sum = foldr (+) 0

or :: [Bool] -> Bool
or = foldr (||) False

We can think of foldr non-recursively as replacing the “cons” operator (:) with the provided binary function, and the empty list with the starting value. For example

foldr (+) 0 [1, 2, 3, 4]
== -- expanding the sugar for the list
foldr (+) 0 (1 : (2 : (3 : (4 : []))))
== -- replacing : with + and [] with 0
(1 + (2 + (3 + (4 + 0))))
==
10

There is also a similar function foldl (“fold left”)

foldl :: (b -> a -> b) -> b -> [a] -> b

which folds from the end of the list, rather than the beginning

foldl (+) 0 [1, 2, 3, 4]
== -- expanding the sugar for the list
foldl (+) 0 (1 : (2 : (3 : (4 : []))))
== -- replacing : with + and [] with 0
((((1 + 2) + 3) + 4) + 0)
==
10

This matches a pattern where the operator associates to the left, rather than the right:

f v [] = v
f v (x:xs) = f (v `op` x) xs

Which maps the empty list to an accumulator value, and any other list to the result of processing the tail using a new accumulator obtained by appling op to the accumulator and the head of the list.

Question

If presented with an infinite list which you wish to process in this manner, should you use foldr or foldl, or does it not matter? Explain your reasoning.

Library functions with folds #

Exercise

Let’s implement a few library functions on lists by using foldr and foldl to check we understand what’s going on.

The template file for this exercise is code/folds-exercise5.hs.

In particular, we’re going to try and implement

-- Length of a list
length' :: [a] -> Int
-- Map a function over a list to produce a new one
map' :: (a -> b) -> [a] -> [b]
-- Reverse a list
reverse' :: [a] -> [a]
-- Check if all entries in a boolean list are true
and' :: [Bool] -> Bool

Using both foldl and foldr.

Higher order functions with folds #

Having done that, we’ll do a few more, this time higher-order functions (again, using the same template file).

Exercise

There are three here that are relatively straightforward, and then one which is much harder (so don’t worry if it eludes you!). Try and implement these using composition of existing higher-order functions that you already know about.

-- Every entry satisfies a predicate, all p [] = True
all' :: (a -> Bool) -> [a] -> Bool
-- At least one entry satisfies a predicate, any p [] = False
any' :: (a -> Bool) -> [a] -> Bool
-- Return elements from a list while they satisfy a predicate, and
-- stop as soon as the predicate is not satisfied
takeWhile' :: (a -> Bool) -> [a] -> [a]
-- Remove elements from a list while they satisfy some predicate, and
-- then return the remainder. This one is MUCH HARDER.
dropWhile' :: (a -> Bool) -> [a] -> [a]

Fold-map fusion #

You might have, and I did, implement all' like this

all' p xs = and (map p) xs
-- Or
all' p = and . map p

This is a nice solution, but it does traverse the list twice, once to apply p to every element, and then to check if they are all True. Fortunately, we can avoid this double traversal, using an example of provably safe program transformations. The particular one we need is fold-map fusion (see Equation 11).

Exercise

Try writing a function

allSingleFold :: (a -> Bool) -> [a] -> Bool
allSingleFold p xs = undefined

which does the same thing as all' but with only one traversal over the list using foldr.

Think about what the operator you want to apply at every stage is.

Hint

The paper linked above shows an equation for fold-map fusion, namely

foldr op z . map p == foldr newop z
  where x `newop` y = p x `op` y

This is a nice property, because it means we can always switch between one form and the other, so optimising compilers can exploit the property.