A benchmark comparing building a string through repeated concatenation vs using an iolist and `Enum.reverse/1` + `:erlang.iolist_to_binary/1`

iolist_benchmark.exs

defmodule IOList do
  @base62_alphabet "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789" |> String.graphemes()

  def random_chars(n) do
    Enum.take_random(@base62_alphabet, n) |> Enum.join()
  end

  def with_iolist(input), do: with_iolist(input, [])
  def with_iolist("", acc), do: acc |> Enum.reverse() |> :erlang.iolist_to_binary()
  def with_iolist(<<first::utf8, rest::binary>>, acc) do
    with_iolist(rest, [first | acc])
  end

  def without_iolist(input), do: without_iolist(input, "")
  def without_iolist("", acc), do: acc
  def without_iolist(<<first::utf8, rest::binary>>, acc) do
    without_iolist(rest, acc <> <<first::utf8>>)
  end
end

inputs = %{
  "small" => IOList.random_chars(1000),
  "medium" => IOList.random_chars(10000),
}

Benchee.run(
  %{
    "with iolist" => fn i -> IOList.with_iolist(i) end,
    "without iolist" => fn i -> IOList.without_iolist(i) end,
  },
  inputs: inputs
)

results.txt

Erlang/OTP 22 [erts-10.5.6] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:1] [hipe]

Operating System: macOS
CPU Information: Intel(R) Core(TM) i7-8559U CPU @ 2.70GHz
Number of Available Cores: 8
Available memory: 16 GB
Elixir 1.9.2
Erlang 22.1.8

Benchmark suite executing with the following configuration:
warmup: 2 s
time: 5 s
memory time: 0 ns
parallel: 1
inputs: medium, small
Estimated total run time: 28 s

Benchmarking with iolist with input medium...
Benchmarking with iolist with input small...
Benchmarking without iolist with input medium...
Benchmarking without iolist with input small...

##### With input medium #####
Name                     ips        average  deviation         median         99th %
with iolist         880.24 K        1.14 μs  ±1536.80%        0.99 μs        1.99 μs
without iolist      349.90 K        2.86 μs   ±744.25%        1.99 μs        5.99 μs

Comparison: 
with iolist         880.24 K
without iolist      349.90 K - 2.52x slower +1.72 μs

##### With input small #####
Name                     ips        average  deviation         median         99th %
with iolist         881.35 K        1.13 μs  ±1539.25%        0.99 μs        1.99 μs
without iolist      351.61 K        2.84 μs   ±793.97%        1.99 μs        5.99 μs

Comparison: 
with iolist         881.35 K
without iolist      351.61 K - 2.51x slower +1.71 μs