Posted on July 31, 2017 by Amir Barylko

Immutability and loops

Working with immutable data structures can be a challenge when transitioning from languages like Java, Javascript or C# (where having a state or modifying the state is a common practice and idiomatic to the language) to languages like Clojure, Haskell, F#, Elixir, etc.

But what does it mean to be immutable? It means the all the code you write or use can not modify an instance of an object or data structure but create a new one that reflects the change you expected.

For example in Clojure to add a new element to a collection you could use conj

(conj [1 2 3 4] 5)

The conj operation does not modify the collection but return a copy of it instead

(let [original [1 2 3 4] copy (conj original 5)]
  (str "The original " original " -- The copy " copy))

Collections in languages that enforce immutability are optimized for these kind of operations so they are very efficient and there is no penalty for immutability.

Friendly recursion

Immutability forces us to rethink how to write loops. Iterating over a collection or creating collections in a loop has to be relearned.

Lets consider transforming the elements of a collection one at a time by applying a function to each element and returning a new collection with the results.

Though we will not use a classic while loop we can do something very similar by using recursion.

Here we see an example of incrementing the elements of a collection one at a time:

(loop [[x & xs] [1 2 3 4] result []]
  (if (nil? x)
    (recur xs (conj result (inc x)))))

Actually this operation is so common that there is a for function that does most of the work for us. This is usually called list comprehension in many languages.

(for [x [1 2 3 4]] (inc x))

Filtering elements given a predicate is also straightforward. Here is an example filtering even elements:

(for [x [1 2 3 4] :when (even? x)] x)

Here is another example for list comprehension taking the first ten elements:

(for [x (range 1 20) :while (< x 11)] x)

Avoid custom loops

List comprehension help us to avoid repeating common operations.

Many of these concepts are indeed so common that specific functions are included in each language to achieve the same functionality.

Transforming elements using a function is called mapping. Thus the function is called map.

(map inc (range 1 20))

This idiom is so popular that many languages implement it. It is very similar in Haskell and F#

-- Haskell
map (+ 1) [1..20]
// F# ((+) 1) [1; 2; 3; 4]

and even you can find it in Java and C# (though is called Select)

// Java Part of `Stream` API
Arrays.asList(1, 2, 3, 4, 5).stream().map(i -> i + 1);
// C# Using LINQ extension methods
new List<int>() { 1, 2, 3, 4, 5 }.Select(i => i + 1);

Why bother?

You may we wondering

Why use other functions instead of just list comprehensions or even a custom loop?

Because functions like map have well defined clear meaning.

Custom loops force the reader to invest effort into looking at the code, writing tests, etc to understand what was the intention behind the loop.

On the other hand, when using well known functions we are applying clearly defined building blocks that we do not need to explain nor document. All that effort comes as a boon and makes our code more idiomatic.

Combining blocks

Using common building blocks make code more expressive and much easier to understand.

The more familiar we are with these blocks the easier is for us to combine them to satisfy our needs.

Here is an example of finding the first even element in a collection:

(first (filter even? [5 3 19 21 6 18 3 4]))

More idiomatic to Clojure here is the same code using the thread last macro.

(->> [5 3 19 21 6 18 3 4]
     (filter even?)

Here is another example taking all the numbers that are greater than zero, incrementing them by one and filtering the even ones.

(->> [5 3 18 21 -1 18 3 4]
     (take-while #(> % 0))
     (map inc)
     (filter even?))

Though there is a combination of functions each element is only traversed once and on demand due to the lazy nature of sequences in Clojure.

Food for thought

I can not emphasize enough how important is to learn as much as we can which common concepts are already implemented and how to use those functions in our code.

Here is a small list as a starting point.

Concatenation and mapping

We already talked about two of the most used functions filter and map. Mapping to a collection and concatenate the result after is a very common scenario. So common that there is a function that composes both.

(mapcat reverse [[3 2 1 0] [6 5 4] [9 8 7]])

Take and drop

Another common scenario is taking (and dropping) elements.

(take 10 (range))

(take-while #(< % 10) (range))

(drop 10 (range 1 20))

(drop-while #(< % 10) (range 1 20))

Zipping collections

So far we looked at working with one element at a time. Here is a very useful function to combine elements of two or more collections.

In Haskell the function is called zip and takes two collections and returns a collection of tuples with one element of each collection.

zip [1, 2, 3] [4, 5, 6] -- Returns [(1, 4), (2, 5), (3, 6)]

zipWith (+) [1, 2, 3] [4, 5, 6] -- Returns [5, 7, 9]

unzip [(1, 4), (2, 5), (3, 6)] -- Returns ([1,2,3],[4,5,6])

-- For more than two collections there is zip3, zip4 etc...

zip3 [1, 2, 3] [4, 5, 6] [7, 8, 9] -- return triplets [(1, 4, 7), ....]

Luckily for us (being a dynamic language) in Clojure to implement the same functionality we use the same map function that we saw in the first example.

The map function expects a function that the arity matches the collections passed as parameter. For two collections it should be a function that receives two parameters, for three collections a function that receives three parameters and son on.

(map vector [1 2 3] [4 5 6] [7 8 9])

(map + [1 2 3] [4 5 6])

Reducing (folding)

Reducing a collection means to transform all the elements not only one element at a time but also including an accumulator that can help us decide what to do.

The reduce function takes first a function with two parameters (the accumulator and the current element), then an initial value and last the collection which to take the elements from.

For example suming all the elements of the collection

(reduce + 0 [1 2 3 4 5 6])

Here is another example converting a collection into a hashmap

(reduce (fn [h [k v]] (assoc h k v)) {} [[:a 1] [:b 2] [:c 3]])

The intermediate steps can be obtained using reductions

(reductions (fn [h [k v]] (assoc h k v)) {} [[:a 1] [:b 2] [:c 3]])

group-by is an especial reduction that returns a hashmap using the function to determine the key of each element.

(group-by even? (range 1 20))

frequencies is a reduction that checks the frequency of each element and returns a hashmap with the element as key and the count as value.

(frequencies [2 2 1 4 4 1 4 2 3])

Extra icing

Here are two more functions that can be useful to have in your toolbox.

partition takes a number and a collection and returns that amount of elements at a time.

(partition 2 (range 20))

And last but not least, here is a very useful function that mimics a scenarios where the same function is applied again and again over the result of the previous call.

And the sequence returned is x, (f x), (f (f x)), (f (f (f x))) and so on.

Because is an infinite sequence we are taking the first twenty memebers of the sequence.

(take 20 (iterate inc 0))

(def powers-of-two (iterate (partial * 2) 1))

(take 10 powers-of-two)


Writing custom loops is a lot of effort and more often than not there is an alternative already written that can help you out. Sometimes needs to be combined with other functions and sometimes requires small helper functions.

Though list comprehensions and many of the functions we just reviewed are very common in functional languages they also exists in many others.

Do not get discouraged, search for them, you may be surprised of what you will find.