Traverse nested list and if the list has two elements apply function

Questions / Help
, ,
1

Hi!

I have been dealing with some problem where I have a nested list, I want to keep the nested structure, but if the list is a two elements list, then i want to apply a custom function (I’m dealing with multypolygon coordinates and want to apply some transform to the coordinates’ system)

So far, I came up with something using recursion:

defmodule NiceModule do
  def apply_function(nested_list) do
    update_in(nested_list, [Access.all()], fn item ->
      case item do
        [x, y] when is_number(x) and is_number(y) -> fun_fun(x, y)  
        sublist when is_list(sublist) -> apply_function(sublist)  
        other -> other  
      end
    end)
  end
  def fun_fun(x,y) do
    {a,b} ={ x+3 , y+2}
    [a,b]
  end
end

aa = [
  [2, 5.0 , [ 1 , 1 ]],
  [[13.0, 17.0], [25.0, 0.0], [0, 0]] ,
  [ 2 ,2 ]
]

NiceModule.apply_function(aa)

Which I think works, but I’m wondering if I’m doing something anti-elixir or if there is a better way (needless to say, I’m new)

Thanks for your time and answers :slight_smile:

2

If that’s all you need then why not simply use Enum.map/2 instead?

    Enum.map(nested_list, fn
      [x, y] when is_number(x) and is_number(y) -> fun_fun(x, y)  
      sublist when is_list(sublist) -> apply_function(sublist)  
      other -> other  
    end)

Creating a tuple that is not needed is just waste of time and resources.

defmodule NiceModule do
  def apply_function(nested_list) do
    Enum.map(nested_list, fn
      [x, y] when is_number(x) and is_number(y) -> [x + 3, y + 2]
      sublist when is_list(sublist) -> apply_function(sublist)  
      other -> other  
    end)
  end
end

This code is 40% (6 of 16 LOC) shorter and therefore definitely more readable.

2 Likes
3

Thanks!

I think this helps clarify things. The fun_fun function was just an example, as what I need is a more complex function. But your clarification helps. If I may ask another question, what might be the use case of Access.all vs Enum.map ?

4

Both of course are valid - they are equivalents in this example. Enum.map/1 is just much simpler here. You use update_in/3 together with Access.all/0 in case you have to update a nested (non-recursive) structure, for example:

defmodule NiceModule do
  def apply_function(map_with_nested_list) do
    update_in(map_with_nested_list, [:aa, Access.all()], &do_apply_function/1)
  end

  defp do_apply_function([x, y]) when is_number(x) and is_number(y), do: [x + 3, y + 2]
  # Pay attention that apply_function/1 call here was changed to do_apply_function1
  defp do_apply_function(sublist) when is_list(sublist), do: do_apply_function(sublist)  
  defp do_apply_function(other), do: other 
end

map = %{
  # aa in your example becomes a key in map
  aa: [
    [2, 5.0 , [ 1 , 1 ]],
    [[13.0, 17.0], [25.0, 0.0], [0, 0]] ,
    [ 2 ,2 ]
  ]
}

NiceModule.apply_function(map)

However if in above example we may use the value of a map key as same as sublist, so we can use Map.update!/3 instead.

defmodule NiceModule do
  def apply_function(map_with_nested_list) do
    Map.update!(map_with_nested_list, :aa, fn
      # the difference is that specified map value can be a list with 2 numbers (not nested)
      [x, y] when is_number(x) and is_number(y) -> [x + 3, y + 2]
      sublist when is_list(sublist) -> apply_function(sublist)
      # similar case here - we allow everything else as a map value
      other -> other  
    end)
  end
end

but nested data does not need to be that simple (nested list or map with nested list), but be more complex. Simply try to find the API that do what you expect in simplest form.

In your example you did not support root list to not be nested or be something else as the item variable is the item in the root list and not the root list itself. To do same in my first example we would need to move pattern-matching to apply_function/1 definition, for example:

defmodule NiceModule do
  # the difference is that specified map value can be a list with 2 numbers (not nested)
  def apply_function([x, y]) when is_number(x) and is_number(y), do: [x + 3, y + 2]
  def apply_function(sublist) when is_list(sublist), do: Enum.map(sublist, &apply_function/1)  
  # similar case here - we allow everything else as a map value
  def apply_function(other), do: other 
end
2 Likes
5

+1 for the other suggestion about using Enum.map if you’re doing a simple “loop over every element in a list”

Another small nitpick: if you’re representing coordinates, an {x, y} tuple would usually be preferred over a two-element list [x, y]. It’s very slightly more space-efficient, and very very very slightly faster to access y.

That could also make your recursion simpler, since then the rule is “if it’s a tuple, transform it; otherwise recurse”.

2 Likes
6

Agree, but there is no “standard” for storing coordinates. In some cases we expect map, struct or tuple - from all of those tuple is indeed the fastest. Anyway I agree that list is used very rarely for such a use case. I did not proposed this as we have no idea from where input comes as in some cases it’s not up to developer to decide about input format.

defmodule NiceModule do
  # skip function guards as we expect every 2-element tuple to always contain numbers
  def apply_function({x, y}), do: {x + 3, y + 2}
  def apply_function(list) when is_list(list), do: Enum.map(list, &apply_function/1)
  def apply_function(other), do: other 
end

or

defmodule NiceModule do
  def apply_function(nested_list) do
    Enum.map(nested_list, fn
      # skip function guards as we expect every 2-element tuple to always contain numbers
      {x, y} -> {x + 3, y + 2}
      sublist when is_list(sublist) -> apply_function(sublist)  
      other -> other  
    end)
  end
end

in case we require list as input.

Going back to topic there is no such thing as “anti-Elixir”. While following so-called community standards or common patterns is much better seen (especially when applying for job) Elixir itself gives you freedom. Even if something is called framework in our ecosystem it’s still library. For example you can store Some.Module where you want - there is no fixed naming and compilation errors. There is no magic check happening in background (except helpers like type checking which simply checks if the code is going to fail).