I’m quite happy with the simplicity of my solution for day 13. The running time of ~20sec makes me somewhat unhappy, although the desired delay is quite large. I wonder how much would I gain by replacing streams with recursion, but don’t have the time to try it out 
sasajuric
Author of Elixir In Action
1
NobbZ
Thanks @sasajuric, based on your code, I was able to get down to 6 - 8 seconds on my office VM, and 4.5 to 5.5 at home…
An important step, additionaly to use your way to represent the firewall, was to replace s |> Enum.take_while(¬_empty?/1) |> Enum.count() by s |> Enum.reduce_while(0, &(if not_empty(&1), do: {:halt, &2}, else: {:cont, &2 + 1})) (rough transcriptions of actual code)
Just using take_while/2 and count/1 were even slower on my office VM than my previous code.
0
sasajuric
Author of Elixir In Action
That’s strange. I just tried out your solution locally, and I get no performance improvement. In other words, I observe no significant perf difference between Stream.take_while |> Enum.count and Enum.reduce_while. In both cases, I get about 20sec of execution time.
0
NobbZ
Thats really funny…
For me (only tested in office) applying the following diff, calculation took ~34-36 seconds, 5 times in a row… Before the patch it was in the range of 6 to 8…
diff --git a/lib/aoc/11-15/day13.ex b/lib/aoc/11-15/day13.ex
index e8c46d2..e016d4c 100644
--- a/lib/aoc/11-15/day13.ex
+++ b/lib/aoc/11-15/day13.ex
@@ -22,9 +22,8 @@ defmodule AoC.Day13 do
0
|> Stream.iterate(&(&1 + 1))
|> Stream.transform(prepare(input), fn n, fw -> {[severities(fw, n)], fw} end)
- |> Enum.reduce_while(0, fn sevs, acc ->
- if Enum.empty?(sevs), do: {:halt, acc}, else: {:cont, acc + 1}
- end)
+ |> Enum.take_while(&(not Enum.empty?(&1)))
+ |> Enum.count()
end
defp prepare(input) do
0
sasajuric
Author of Elixir In Action
You have Enum.take_while, in my solution I used Stream.take_while, so that might explain it.
0
NobbZ
Yeah, I realized it right after submitting, with Stream.take_while/2, its ~7 seconds through 3 runs now.
0
sasajuric
Author of Elixir In Action
Yeah, that makes much more sense, since with Enum you generate a huge list (my result was close to 4 millions).
0
NobbZ
sasajuric
Author of Elixir In Action
Using this monstrosity, I was able to shave it off by some 40%. The code is quite ugly, but the idea is fairly simple, based on sieves. I’m generating failed positions in sorted order, and then stop when the difference between two consecutive positions is > 1. Not sure if the code is 100% correct, but it gives the same output for my input 
2
