Recursion
Mix.install([
{:kino, github: "livebook-dev/kino", override: true},
{:kino_lab, "~> 0.1.0-dev", github: "jonatanklosko/kino_lab"},
{:vega_lite, "~> 0.1.4"},
{:kino_vega_lite, "~> 0.1.1"},
{:benchee, "~> 0.1"},
{:ecto, "~> 3.7"},
{:math, "~> 0.7.0"},
{:faker, "~> 0.17.0"},
{:utils, path: "#{__DIR__}/../utils"}
])
Navigation
Setup
Ensure you type the ea
keyboard shortcut to evaluate all Elixir cells before starting. Alternatively you can evaluate the Elixir cells as you read.
Recursion
Recursion is a function that calls itself.
flowchart LR
b[Function] --> a[Function]
a --> b
Recursion creates a loop effect where the function calls itself over and over.
defmodule Recursion do
def loop(n) do
if n > 0 do
IO.puts(n)
loop(n - 1)
end
end
end
Recursion.loop(5)
This calls the loop function 5 times. Under the hood, this places 5 stack frames on the stack.
We trigger a call to IO.puts/2
to show that the loop function has been called 5 times with
a different argument.
flowchart LR
a["loop(n)"]
b["loop(n - 1)"]
c["loop(n - 2)"]
d["loop(n - 3)"]
e["loop(n - 4)"]
f["loop(...)"]
5["loop(5)"]
4["loop(4)"]
3["loop(3)"]
2["loop(2)"]
1["loop(1)"]
5 --> 4 --> 3 --> 2 --> 1
a --> b --> c --> d --> e --> f
Computerphile explains recursion in excellent detail.
Kino.YouTube.new("https://www.youtube.com/watch?v=Mv9NEXX1VHc")
Endless Recursion
We should have some end condition, otherwise, this would run forever. You’ll notice that
this Elixir cell never stops running. Under the hood it’s calling Forever.run/0
over and over.
defmodule Forever do
def run do
run()
end
end
Forever.run()
Stack Overflow
Coming from another language, you might be surprised that the endless recursion function doesn’t crash in Elixir. In most programming languages, calling a recursive function puts too many stack frames on the stack, and causes a stack overflow.
That’s because stack memory gets too full (overflowed) storing each stack frame of the recursive call.
Tail Recursion and Tail-Call Optimization
Since functional programming languages rely so much on recursion, Elixir (and Erlang) implement tail-call optimization.
Tail-call optimization circumvents adding new stack frames, instead, it reuses the current stack frame and jumps back to the top of the stack frame. This avoids additional memory consumption.
Body Recursion
Keep in mind that Elixir can only tail-call optimize your recursive function if the last thing it does is call itself. That’s why it’s called tail recursion. Otherwise, if the function calls itself in the body, it’s called body-recursion and is not optimized.
Your Turn
Let’s prove that body-recursion is not optimized. First, open the runtime panel in this livebook.
Press s
then r
to open the settings panel. There you can see the current memory consumption.
Next, uncomment and execute the following Elixir cell that uses body recursion. It’s a nonsense function that doesn’t do anything, however, it will infinitely call itself in the body of the function.
You’ll notice the Process memory consumption will increase, and eventually, the cell will abort. You may need to click the Connect button to reconnect the Elixir runtime.
# defmodule Body do
# def recursion([head | tail]) do
# recursion(tail ++ tail) ++ [head]
# end
# end
# Body.recursion([1,2,3])
Make sure you comment out the code above, otherwise Livebook will keep disconnecting.
Using Recursion
So why is recursion useful? Well, it’s how we achieve a great deal of functionality in Elixir.
For example, many functions in the Enum
module use recursion under the hood.
Recursion allows us to accomplish enumeration.
For example, we could use recursion to enumerate through and sum the elements in a list.
defmodule Recursion do
def sum(list, accumulator \\ 0) do
case list do
[] -> accumulator
[head | tail] -> sum(tail, accumulator + head)
end
end
end
Recursion.sum([1, 2, 3], 0)
We enumerate through the list by recursively calling sum/2
on the tail of the list and building
an accumulator. In this case, the initial accumulator is 0
.
Each element in the list adds to the accumulator. 1
+ 2
+ 3
= 6
so the function returns 6
.
flowchart LR
sum1["sum([1, 2, 3], 0)"]
sum2["sum([2, 3], 1)"]
sum3["sum([3], 3)"]
sum4["sum([], 6)"]
sum1 --> sum2 --> sum3 --> sum4 --> 6
Walking through the function we bind list
to [1, 2, 3]
and accumulator to 0.
def sum(list, accumulator) do
case list do
# [] -> accumulator
[head | tail] -> sum(tail, accumulator + head)
end
end
list
is not empty so the program recursively calls sum([2, 3], 0 + 1)
.
def sum([1, 2, 3], 0) do
case [1, 2, 3] do
# [] -> accumulator
[1 | [2, 3]] -> sum([2, 3], 0 + 1)
end
end
list
is still not empty, so the program recursively calls sum([3], 1 + 2)
.
def sum([2, 3], 1) do
case [2, 3] do
# [] -> accumulator
[2 | [3]] -> sum([3], 1 + 2)
end
end
list
is still not empty, so the program recursively calls sum([], 3 + 3)
.
def sum([3], 3) do
case [3] do
# [] -> accumulator
[3 | []] -> sum([], 3 + 3)
end
end
list
is empty, so the program returns the accumulator (6
).
def sum([], 6) do
case [] do
[] -> 6
# [head | tail] -> sum(tail, accumulator + head)
end
end
Rather than using case statements, it’s common to use different function clauses to handle the base case. The base case is when the function returns a value instead of making a recursive call.
In the Recursion
module, the base case is when the list
is empty and sum/2
returns the accumulator
.
With multiple function clauses, the Recursion
module could instead be implemented like so:
defmodule Recursion do
def sum([], accumulator), do: accumulator
def sum([head | tail], accumulator), do: sum(tail, accumulator + head)
end
Recursion.sum([1, 2, 3], 0)
Your Turn
In the Elixir cell below, implement a CustomEnum.map/2
function for lists using recursion.
Do not use the Enum
module.
CustomEnum.map([1, 2, 3], fn each -> each * 2 end)
[2, 4, 6]
defmodule CustomEnum do
def map(list, function) do
end
end
Utils.feedback(:custom_enum_map, CustomEnum)
Commit Your Progress
Run the following in your command line from the project folder to track and save your progress in a Git commit.
$ git add .
$ git commit -m "finish recursion section"