IO lists with a comma or a bar

Questions / Help
1

I’ve been reading a blog post [0] about IO lists and noticed that the authors build them with commas

names = ["A", "B"]    # => ["A", "B"]
names = [names, "C"]  # => [["A", "B"], "C"]

And I usually see | operator used to build lists. And since the lists built with a comma and the lists built with a bar are not the same to the VM,

iex(14)> names = ["A", "B"]
["A", "B"]
iex(15)> [names | "C"] == [names, "C"]
false

I wonder what the difference between them actually is.

[0] https://www.bignerdranch.com/blog/elixir-and-io-lists-part-1-building-output-efficiently/

2

The output from IO.puts is the same for both [names | "C"] and [names, "C"] though.

3

Given your example

names = ["A", "B"]
  • [names | "C"] # => [["A", "B"] | "C"]. This is an improper list.
  • [names, "C"] # => [["A", "B"], "C"]. This is a list where the first element is a list (["A", "B"]), and the second element is a string ("C").

In order to use | to create a proper list, you will need to have the list variable as the following element.

["C" | names] => ["C", "A", "B"]

4

Are there any performance implications of using one vs the other?

5

Depends on your definition of performance. If you are talking strictly speed and memory, I do not think so. With that in mind:

iex(1)> length [1,2,3]
3
iex(2)> length [1,2,3 | 4]
** (ArgumentError) argument error
    :erlang.length([1, 2, 3 | 4])

I do not know how many of the standard library functions actually work with improper lists.

6

Well, performance regarding IO operations, like writing to a socket, as in the blog post that I’ve linked to …

7

I’ve made a simple test with benchfella

defmodule IOListBench do
  use Benchfella
  
  bench "proper", [list: get_list()] do
    [list, "C"]
  end

  bench "improper", [list: get_list()] do
    [list | "C"]
  end

  def get_list do
    for _ <- 1..100_000, do: "asd"
  end
end

and the results are the same.

8

IO operations are perfectly fine with improper lists. Using improper lists for concatenating iodata is slightly more efficient in terms of memory.

iex(1)> ab = ["A", "B"]
["A", "B"]
iex(2)> [ab | "C"]
[["A", "B"] | "C"]
iex(3)> [ab, "C"]
[["A", "B"], "C"]
iex(4)> [ab | ["C"]]
[["A", "B"], "C"]
iex(5)> :erts_debug.flat_size([ab, "C"])
26
iex(6)> :erts_debug.flat_size([ab | "C"])
24

When you do [ab, "C"] the “raw” representation would be [ab | ["C" | []]], whereas for [ab | "C"] that’s all there is - there’s one less list “cell” involved.

2 Likes