EFLFE: Elixir Flavoured Lisp Flavoured Erlang

History

In 2019 I was invited to give a talk at ElixirConfLA, at that point I didn't know Elixir so I decided to "make a joke" and instead of learning Elixir I would create a transpiler from Erlang to Elixir.

The Proof of Concept as a Joke was a lot of work but at least I learned a lot about Elixir and Pretty Printers.

One year later I was invited to CodeBEAM Brasil and I decided to push the PoC to completion to achieve the goal of transpiling Erlang/OTP and the transpiler itself.

This year I was invited again and I felt the pressure to continue with the tradition.

Last year my talk was with Robert Virding Co-creator of Erlang and creator of LFE (Lisp Flavoured Erlang), I had the idea to transpile LFE too.

At that point LFE 1.0 compiled to an internal representation (Core Erlang) that was one level below the one I was using in Elixir Flavoured Erlang.

I knew that the next major version of LFE was going to switch to the representation I was using, so it was just a matter of waiting for the release.

Timing helped and LFE 2.0 was released in June.

I went code diving and found how to get the data at the stage I needed and then fixing some corner cases around naming (LFE uses lispy Kebab case).

Result

The result is EFLFE: Elixir Flavoured Lisp Flavoured Erlang an Lisp Flavoured Erlang to Elixir transpiler.

Business in the Front

Aliens in the Back

Run it with a configuration file and a list of LFE files:

./efe pp-lfe file.conf my-code.lfe

And it will output Elixir files for each.

Example

Input:

(defmodule ping_pong
  (export
    (start_link 0)
    (ping 0))
  (export
    (init 1)
    (handle_call 3)
    (handle_cast 2)
    (handle_info 2)
    (terminate 2)
    (code_change 3))
  (behaviour gen_server))        ; Just indicates intent

(defun start_link ()
  (gen_server:start_link
    #(local ping_pong) 'ping_pong '() '()))

;; Client API

(defun ping ()
  (gen_server:call 'ping_pong 'ping))

;; Gen_server callbacks

(defrecord state
  (pings 0))

(defun init (args)
  `#(ok ,(make-state pings 0)))

(defun handle_call (req from state)
  (let* ((new-count (+ (state-pings state) 1))
         (new-state (set-state-pings state new-count)))
    `#(reply #(pong ,new-count) ,new-state)))

(defun handle_cast (msg state)
  `#(noreply ,state))

(defun handle_info (info state)
  `#(noreply ,state))

(defun terminate (reason state)
  'ok)

(defun code_change (old-vers state extra)
  `#(ok ,state))

Result:

defmodule :ping_pong do
  use Bitwise
  @behaviour :gen_server
  def start_link() do
    :gen_server.start_link({:local, :ping_pong}, :ping_pong, [], [])
  end

  def ping() do
    :gen_server.call(:ping_pong, :ping)
  end

  require Record
  Record.defrecord(:r_state, :state, pings: 0)

  def init(args_0) do
    {:ok, r_state(pings: 0)}
  end

  def handle_call(req_0, from_0, state_0) do
    new_count_0 = r_state(state_0, :pings) + 1

    (
      new_state_0 = r_state(state_0, pings: new_count_0)
      {:reply, {:pong, new_count_0}, new_state_0}
    )
  end

  def handle_cast(msg_0, state_0) do
    {:noreply, state_0}
  end

  def handle_info(info_0, state_0) do
    {:noreply, state_0}
  end

  def terminate(reason_0, state_0) do
    :ok
  end

  def code_change(old_vers_0, state_0, extra_0) do
    {:ok, state_0}
  end

  def unquote(:"LFE-EXPAND-EXPORTED-MACRO")(_, _, _) do
    :no
  end
end

Where The Code Gets Ugly

LFE and Elixir both share the fact that they support macros, macros are expanded at compile time and are a language feature, that means that when I get the code to transpile it, it's already expanded.

If the module you are transpiling uses macros you will transpile the macro expanded version of the code, which may be okay or not depending on the kind of code that the macro generates.

Remaining Work

The remaining work is to understand the details of variable scoping in LFE and see if it's compatible with Elixir so that I can translate it as is like I'm doing now.

If they differ I have to see if I can do some local analysis to transform it so that the resulting code behaves semantically like the original.

If you try it and have some questions let me know at @warianoguerra or in the repo's issue tracker.