Help composing/abstracting list of keys to traverse nested data structures using the Access module

Hello, I’m looking to dry up some LiveView handle_info/event callbacks that require modifying a nested data structure using update_in/3 in response to CRUD, or rather CUD, actions. The second parameter for update_in/3 is a list of keys that are often functions from the Access module e.g. Access.key!/2 and Access.filter/2.

For the sake of a concrete example, Person has many Pets, Pet has many Toys, and there are callbacks that update a preloaded person struct assigned to the socket after a user adds, updates, and/or deletes a toy. For the add and delete callbacks, update_in/3 only has to reach into the toys list whereas the update callback goes a step further to find the specific toy within in the toy list.

How would I go about abstracting out the common parts of the traverse path?

person =
  %Person{
    pets: [
      %Pet{
        toys: [
          %Toy{}
        ]
      }
    ]
  }
# within toy created callback
update_in(
  person,
  [
    Access.key!(:pets),
    Access.filter(&(&1.id == pet_id)),
    Access.key!(:toys)
  ],
  fn toys -> [toy | toys] end
) 

# within toy deleted callback
update_in(
  person,
  [
    Access.key!(:pets),
    Access.filter(&(&1.id == pet_id)),
    Access.key!(:toys)
  ],
  fn toys -> Enum.reject(toys, &(&1.id == toy_id)) end
) 

# within toy updated callback
update_in(
  person,
  [
    Access.key!(:pets),
    Access.filter(&(&1.id == pet_id)),
    Access.key!(:toys),
    Access.filter(&(&1.id == toy_id))
  ],
  fn toy -> %{toy | new_attributes} end
) 
# composing list of keys within toy updated callback
update_in(
  person,
  [
    Access.key!(:pets),
    Access.filter(&(&1.id == pet_id)),
    Access.key!(:toys)
  ] ++ [Access.filter(&(&1.id == toy_id))],
  fn toy -> %{toy | new_attributes} end
) 

As demonstrated above, the first three elements in the second parameter list of keys are shared across all three callbacks. Is there a sensible way to refactor and abstract that out? I imagine some form of macros/quote/unquote would be involved… would the added brevity even be worth it given the increased complexity?

update_in/3 and friends are plain functions, and their argument is a plain list (of anonymous functions). The macro-sorcery might be necessary to do things with update_in/2, but not here.

You could do this with functions:

defp toys_for(pet_id) do
  [
    Access.key!(:pets),
    Access.filter(&(&1.id == pet_id)),
    Access.key!(:toys)
  ]
end

# in the created/deleted callbacks:
update_in(
  person,
  toys_for(pet_id),
  fn toys -> [toy | toys] end
) 

# in the updated callback:
update_in(
  person,
  toys_for(pet_id) ++ [Access.filter(&(&1.id == toy_id))],
  fn toy -> %{toy | new_attributes} end
) 

You could also split this differently - combine Access.key!(:pets) with the Access.filter and name it pet_with_id or similar.

3 Likes

Ahh of course, for some reason didn’t occur to just pass in the ids. Anyways, thanks @al2o3cr and thanks Elixir for first class functions!

1 Like

Played around with splitting it like @al2o3cr suggested and got something like this:

  defp access_pet_with_id(id) do
    [
      Access.key!(:pets),
      Access.filter(&(&1.id == id)),
    ]
  end

  defp access_toy_with_id(id) do
    [
      Access.key!(:toys),
      Access.filter(&(&1.id == id)),
    ]
  end

And just for fun, took a crack at generating these formulaic helper functions for any given list of preloaded Ecto structs:

  for struct <- [%Pet{}, %Toy{}] do
    schema = struct.__meta__.schema
      |> Atom.to_string
      |> String.split(".")
      |> List.last
      |> String.downcase

    defp unquote(String.to_atom("access_#{schema}_with_id"))(id) do
      [
        Access.key!(unquote(String.to_atom(schema <> "s"))),
        Access.filter(&(&1.id == id))
      ]
    end
  end

It’s overkill for this toy example, but it might be helpful for others in some situations.

This is probably in the realm of chrome plating, but you could even generalize to

  defp access_key_with_id(key, id) do
    [
      Access.key!(key),
      Access.filter(&match?(^id, &1.id)),
    ]
  end
1 Like

Shiny! That also opens the door for something like this:

defp access_keys_with_ids(list) when is_list(list) do
  Enum.reduce(list, [], fn {key, id}, acc -> 
    acc ++ access_key_with_id(key, id)
  end)
end

access_keys_with_ids([pets: 1, toys: 2])
=> [
       Access.key!(:pets),
       Access.filter(&match?(1, &1.id)),
       Access.key!(:toys),
       Access.filter(&match?(2, &1.id))
   ]