Control structure `with else` problem

I am playing with elixir, phoenix and ecto. And ran into trouble.
Repo.get return nil because Room table is empty when test start. But else block with nil pattern don’t work.
My code:

test "unexisted room" do
    with room <- App.Repo.get(App.Room, 1),
         true <- test_fn("good") do
          IO.inspect(room)
    else
      nil -> IO.puts("Room does not exist")
      false -> IO.puts("Not good")
      _ -> IO.puts("error")
    end
  end

  defp test_fn(good?) do
    case good? do
      "good" -> true
      _ -> false
    end
  end

result mix.test:
nil.

If i change true <- test_fn("some not good") it return false and else block work fine
result mix.test:
Not good.

Then for test i add nil case in test_fn and add call test_fn in with block
Code:

test "unexisted room" do
    with room <- App.Repo.get(App.Room, 1),
         true <- test_fn(nil),
         true <- test_fn("false") do
          IO.inspect(room)
    else
      nil -> IO.puts("Room does not exist")
      false -> IO.puts("Not good")
      _ -> IO.puts("error")
    end
  end

  defp test_fn(good?) do
    case good? do
      "good" -> true
      nil -> nil
      _ -> false
    end
  end

result mix.test:
Room does not exist.

Why with else don’t work with ecto repo?

In your first example,

  • When App.Repo.get(App.Room, 1) returns nil it will still bind nil to room.
  • Then test_fn("good") will return true so that will also pass.
  • Lastly. it will execute IO.inspect(room) and since room == nil you see nil.

From that explanation I think you can work out the second example. You might find some value in with room when not is_nil(room) <- App.Repo.get(App.Room, 1).

1 Like

Thanks, here is my problem

it will still bind nil to room

My solution

with %App.Room{} = room <- App.Repo.get(App.Room, 1),
     true <- test_fn("good") do
       IO.inspect(room)
else
  nil -> IO.puts("Room does not exist")
  false -> IO.puts("Not good")
   _ -> IO.puts("error")
end

If the else behaviour changes depending on the type of failure, it is generally considered clearer to put those behaviours inside of their own functions and then use those functions inside the with statement.

For example, your code can be refactored to this:

with %App.Room{} = room <- get_room(1),
    true <- test_fn("good") do
  IO.inspect(room)
end

defp get_room(room_id) do
  case App.Repo.get(room_id) do
    nil ->
      IO.puts("Room does not exist")
      nil

    room ->
     room
  end
end

defp test_fn(good?) do
  case good? do
   "good" -> 
     true

   _ -> 
    IO.puts("Not good")
    false
  end
end

or this

with {:ok, room} <- get_room(1),
    {:ok, true} <- test_fn("good") do
  IO.inspect(room)
else
  {:error, reason} -> IO.puts(reason)
end

defp get_room(room_id) do
  case App.Repo.get(room_id) do
   nil ->
      {:error, "Room does not exist"}

   room ->
     {:ok, room}
  end
end

defp test_fn(good?) do
    case good? do
      "good" -> 
        {:ok, true}

      _ -> 
        {:error, "Not Good"}
    end
end

Some references:

1 Like