Has anyone else ever wanted to merge two maps, but have the resulting map only include keys common to both maps? I think of it as analogous to MapSet.intersection but with a function for handling the values.

m1 = %{a: 1, b:2}
m2 = %{b: 3, c:4}
Map.intersection(m1, m2, fn _k, v1, v2 -> v1 * v2 end)
> %{b: 6}

This would replace

Map.merge(m1, m2, fn _k , v1, v2 -> v1 * v2 end) 
|> Enum.filter(fn {k,_v} -> [m1, m2] |> Enum.all?(&Map.has_key?(&1, k)) end)
>%{b: 6}


1 Like

All you have to do is a simple reducer. :smiling_imp:

defmodule Example do
  @spec sample(map(), map(), (Map.key(), Map.value(), Map.value() -> Map.value())) :: map()
  def sample(left, right, func) when is_map(left) and is_map(right) and is_function(func, 3) do
    Enum.reduce(left, %{}, fn
      {key, value}, acc when is_map_key(right, key) ->
        Map.put(acc, key, func.(key, value, right[key]))

      _pair, acc ->

iex> Example.sample(%{a: 1, b: 2}, %{b: 3, c: 4}, fn _k, v1, v2 -> v1 * v2 end)
%{b: 6}

Helpful resources:

  1. Guides |> Typespecs @ Elixir documentation
  2. Enum.reduce/3
  3. Kernel.is_function/2
  4. Kernel.is_map/1
  5. Kernel.is_map_key/2
  6. Map.put/3

I think it should be called merge_intersection or something like that? Union means anything in either of the things.


Of course. Intersection not union. Brain fart

You can do MapSet.intersection on the keys of both maps and then Map.take with the result of it.

But I believe @Eiji’s solution will be more performant.

def merge_intersection(map1, map2, fun) do
  for key <- Map.keys(map1) -- Map.keys(map1) -- Map.keys(map2), into: %{} do
    {key, fun.(map1[key], map2[key])}

Sorry, I just hit the wrong “reply” button :sweat_smile:

:maps.intersect(m1, m2)
%{b: 3}

:maps.intersect_with(fn _k , v1, v2 -> v1 * v2 end, m1, m2)
%{b: 6}

I’m assuming that since Elixir’s maps preceded Erlang’s the :maps module is not simply wrapped by Map which is why there is no Map.intersect or Map.intersect_with. Are Elixir maps completely interoperable with the Erlang :maps implementation?

There is only ONE type of map in the BEAM. The maps in Elixir are exactly the same maps as in Erlang. Seeing Erlang is the base language on the BEAM Elixir could not have had maps before the BEAM and Erlang implemented them.

iex(1)> m1 = %{a: 1, b: 1, c: 1}
%{a: 1, b: 1, c: 1}
iex(2)> m2 = %{a: 2 , c: 2, d: 2}
%{a: 2, c: 2, d: 2}
iex(3)> mi = :maps.intersect(m1, m2)
%{a: 2, c: 2}
iex(4)> :io.write(m1)       
#{a => 1,b => 1,c => 1}:ok
iex(5)> :io.write(m2)
#{a => 2,c => 2,d => 2}:ok
iex(6)> :io.write(m1)               
#{a => 1,b => 1,c => 1}:ok

The :io.write function is an output function in the Erlang io module and prints the maps which shows they are the same maps as in Erlang.

The :maps.intersect/2 function first came in OTP 24.0 so it may not be used yet in the Map module.


Thanks for clarifying. I was thinking of MapSets.

Elixir main requires Erlang/OTP 24. Therefore if someone wants to submit a PR that adds intersect/2 and intersect/3, it will be welcome.


Just in case anybody has ideas:


Or maybe HashDict — Elixir v1.14.2?

No just misinterpreting from

So :sets existed before MapSet but MapSet did not build on :sets, but Map always build on :maps?

1 Like


MapSet did not build on :sets because :sets were not backed by maps but now they are.

1 Like