Phoenix and Elm, a real use case (pt. 6)

Phoenix and Elm communication through WebSockets
Mar 20, 2017 · 14 min read
elixir
phoenix
elm

Phoenix and Elm WebSockets support

In the last part, we refactored our application to add Elm routing and added the show contact route. These changes include adding a new API endpoint in the ContactController module to return a given contact's JSON representation. However, since I started using Phoenix, I have found myself using fewer controllers, and more WebSockets and Phoenix's channels. The main reasons for doing this are:

  • They are much faster than Http requests.
  • Once the connection is established, and the authentication is accomplished, you do not need to authenticate future requests and rely on cookies for it.
  • Its PubSub system opens a new world of possibilities.

That being said, today we are going to replace the current API controller we have with a new Phoenix channel and learn how to add support for this new channel in Elm. Let's do this!

Phoenix changes

First of all, let's get rid of what we are not going to need anymore. This includes deleting the ContactContoller file, located in web/controllers/contact_controller.ex and the ContactView in web/views/contact_view.ex. We also need to update the router file to remove the API pipeline:

# web/router.ex

defmodule PhoenixAndElm.Router do
  use PhoenixAndElm.Web, :router

  pipeline :browser do
    plug :accepts, ["html"]
    plug :fetch_session
    plug :fetch_flash
    plug :protect_from_forgery
    plug :put_secure_browser_headers
  end

  scope "/", PhoenixAndElm do
    pipe_through :browser # Use the default browser stack

    get "/*path", PageController, :index
  end
end

Once we have removed the unnecessary stuff, let's create the channel which is going to replace the old ContactController. First, we need to declare it in the UserSocket module:

# web/channels/user_socket.ex

defmodule PhoenixAndElm.UserSocket do
  use Phoenix.Socket

  ## Channels
  channel "contacts", PhoenixAndElm.ContactsChannel

  # ...
end

Next step is to create the ContactsChannel module in the same folder:

# web/channels/contacts_channel.ex

defmodule PhoenixAndElm.ContactsChannel do
  use Phoenix.Channel
  alias PhoenixAndElm.{Contact, Repo}
  import Ecto.Query, only: [order_by: 2]

  require Logger

  def join("contacts", _, socket), do: {:ok, socket}

  def handle_in("contacts:fetch", params, socket) do
    Logger.info "Handling contacts..."

    search = Map.get(params, "search") || ""

    page = Contact
    |> Contact.search(search)
    |> order_by(:first_name)
    |> Repo.paginate(params)

    {:reply, {:ok, page}, socket}
  end

  def handle_in("contact:" <> contact_id, _, socket) do
    Logger.info "Handling contact..."

    contact = Contact
    |> Repo.get(contact_id)

    case contact do
      nil ->
        {:reply, {:error, %{error: "Contact no found"}}, socket}
      _ ->
        {:reply, {:ok, contact}, socket}
    end
  end
end

The ContactsChannel handles two events:

  • contacts:fetch wich handles contact search and pagination like the old index action of the ContactController.
  • contacts:id which returns a given contact's data, just like the old show action.

Depending on the environment the application is running, the socket URL will probably change. The front-end application needs to know the URL to create the connection, so we need it to pass it somehow. The easiest way of doing so is by creating a helper method in a Phoenix view and call it to assign it to javascript globally. Let's add the helper method in the LayoutView module:

# web/views/layout_view.ex

defmodule PhoenixAndElm.LayoutView do
  use PhoenixAndElm.Web, :view

  def socket_url, do: System.get_env("WEBSOCKECT_URL") || "ws://localhost:4000/socket/websocket"
end

The socket_url function returns the value in the WEBSOCKECT_URL system variable, or the default one if not set. Now we can update the main template file to call this function and set value in javascript:

# web/templates/layout/app.html.eex

<!DOCTYPE html>
<html lang="en">
  # ...

  <body>
    # ...

    <script>window.socketUrl = '<%= PhoenixAndElm.LayoutView.socket_url %>';</script>
    <script src="<%= static_path(@conn, "/js/app.js") %>"></script>
  </body>
</html>

Once the socketUrl value is set, we can use it in the app.js file:

// web/static/js/app.js

import Elm from './main';

const elmDiv = document.querySelector('#elm_target');

if (elmDiv) {
  const socketUrl = window.socketUrl;

  Elm.Main.embed(elmDiv, { socketUrl });
}

Cool, but how does Elm handle this values?

Program with flags

Conveniently, Elm has the concept of flags which are values that can be received from JavaScript while creating the application. To use flags, we need to do some refactoring in the application, so let's start by editing the Main module:

-- web/elm/Main.elm

module Main exposing (..)

-- ...

init : Flags -> Navigation.Location -> ( Model, Cmd Msg )
init flags location =
    let
        currentRoute =
            parse location

        model =
            initialModel flags currentRoute
    in
        urlUpdate model


main : Program Flags Model Msg
main =
    Navigation.programWithFlags UrlChange
        { init = init
        , view = view
        , update = update
        , subscriptions = always <| Sub.none
        }

The main function now has a Flags type in its definition, that is passed to the init function and from there to the initialModel. Let's move on to the Model module to implement this type:

-- web/elm/Model.elm

module Model exposing (..)

-- ...

type alias Flags =
   { socketUrl : String }


type alias Model =
   { -- ---
   , flags : Flags
   }


initialModel : Flags -> Route -> Model
initialModel flags route =
   { -- ...
   , flags = flags
   }

The Flags type is a record containing a socketUrl key. As we want to initialize this value with the flags received from javascript, we pass it to the initialModel function. If we now refresh the browser and take a closer look at the debugger history, we can see how the flags key is already set:

Flags

Now we are ready to start implementing the socket communication between the Elm program and the Phoenix backend.

The elm-phoenix package

There are many different ways to add WebSockets support in Elm, but my favorite one so far is using the elm-phoenix package. This package is an effect manager, so at the moment is not in the elm package repository. Therefore, its installation can be tricky depending on your Phoenix configuration, so easiest way I have found is by downloading it and adding the source files into a vendor folder. Before doing this, let's move all of our Elm module files into a new web/elm/src folder. After doing it, it should look like this:

web/elm
├── elm-package.json
├── elm-stuff
└── src
    ├── Commands.elm
    ├── Common
    │   └── View.elm
    ├── Contact
    │   └── View.elm
    ├── ContactList
    │   └── View.elm
    ├── Decoders.elm
    ├── Main.elm
    ├── Messages.elm
    ├── Model.elm
    ├── Routing.elm
    ├── Update.elm
    └── View.elm

Next, create a new web/elm/vendor folder and copy there the elm-phoenix source files:

web/elm
├── elm-package.json
├── elm-stuff
├── src
└── vendor
    ├── Phoenix
    │   ├── Channel.elm
    │   ├── Internal
    │   │   ├── Channel.elm
    │   │   ├── Helpers.elm
    │   │   ├── Message.elm
    │   │   └── Socket.elm
    │   ├── Push.elm
    │   └── Socket.elm
    └── Phoenix.elm

Now we have to make the Elm compiler aware of these changes, so let's update the elm-package.json file:

// web/elm/elm-package.json

{
    "version": "1.0.0",
    "summary": "helpful summary of your project, less than 80 characters",
    "repository": "https://github.com/user/project.git",
    "license": "BSD3",
    "source-directories": [
        "./src",
        "./vendor"
    ],
    // ...
}

We also need to change Brunch's configuration; otherwise, the build is not going to succeed:

// brunch-config.js

exports.config = {
  // ...

  // Configure your plugins
  plugins: {
    // ...

    elmBrunch: {
      elmFolder: 'web/elm',
      mainModules: ['src/Main.elm'],
      outputFolder: '../static/js',
      makeParameters: ['--debug'],
    },

    //...
  }
}

Finally, elm-phoenix depends on the elm-lang/websocket package, so let's install it:

$ cd web/elm && elm-package install elm-lang/websocket -y

After doing this and restarting the Phoenix server, everything should be working as before. In case it does not, here is the commit with all these changes.

The Subscriptions module

In order to listen to external input, such as mouse events or location changes, Elm uses subscriptions. In our case, as we are using elm-phoenix, we want to handle WebSockets events and make the application respond to them. Let's create a new Subscriptions module:

-- web/elm/src/Subscriptions.elm

module Subscriptions exposing (..)

import Messages exposing (Msg(..))
import Model exposing (Model)
import Phoenix
import Phoenix.Channel as Channel exposing (Channel)
import Phoenix.Socket as Socket exposing (Socket)


subscriptions : Model -> Sub Msg
subscriptions model =
    Phoenix.connect (socket model.flags.socketUrl) [ contacts ]


socket : String -> Socket Msg
socket socketUrl =
    Socket.init socketUrl


contacts : Channel Msg
contacts =
    Channel.init "contacts"
        |> Channel.withDebug

The subscriptions function prepares the socket and the channel we want to join, using the program's flags to set the socketUrl. In our case, we only have a single channel, so we do not need to add extra configuration to check which channel or channels it has to join, but if you feel curious about how to join multiple channels, there is an excellent demo example in the official repo. In our case, we are good to continue, so let's update the Main module, so the main program uses the new subscriptions function:

-- web/elm/src/Main.elm

module Main exposing (..)

import Subscriptions exposing (subscriptions)

-- ...

main : Program Flags Model Msg
main =
    Navigation.programWithFlags UrlChange
        { init = init
        , view = view
        , update = update
        , subscriptions = subscriptions
        }

If we refresh the browser and check our browser's console, we can see two log messages, one when the program tries to join the ContactsChannel and another one when the join result is successful:

Socket connection

Commands refactoring

Now that we have the connection to the WebSocket ready, we need to refactor the Commands module to replace the old Http requests with WebSocket ones:

-- web/elm/src/Commands.elm

module Commands exposing (..)

import Phoenix exposing (..)
import Phoenix.Push as Push
-- ...

fetch : String -> Int -> String -> Cmd Msg
fetch socketUrl page search =
    let
        payload =
            JE.object
                [ ( "page", JE.int page )
                , ( "search", JE.string search )
                ]

        push =
            Push.init "contacts" "contacts:fetch"
                |> Push.withPayload payload
                |> Push.onOk FetchSuccess
                |> Push.onError FetchError
    in
        Phoenix.push socketUrl push


fetchContact : String -> Int -> Cmd Msg
fetchContact socketUrl id =
    let
        push =
            Push.init "contacts" ("contact:" ++ toString id)
                |> Push.onOk FetchContactSuccess
                |> Push.onError FetchContactError
    in
        Phoenix.push socketUrl push

The fetch function now receives one more parameter with the socketUrl value that it is going to need to send the request. It first creates a JSON payload with the page and search, just like the old Http one. Next, it creates a push, from the elm-phoenix package, in which passes the Phoenix's channel name and the event, in this case contacts:fetch. It attaches the previously created payload using withPayload and sets the messages for both the onOk and onError callbacks, triggered depending on the result. Finally, it sends the push to the socketUrl.

On the other hand, fetchContact is more simple. It creates a push to the same channel, but for the contact:* where * is the id of the contact we want to fetch. It only sets the messages for the onOk and onError callbacks, and finally sends the push through the socketUrl.

Messages and the Update module

At this point, the compiler must be some errors related to the new messages we have added in the Commands module, but we have not declared it yet. So let's edit the Messages module to fix this:

-- web/elm/src/Messages.elm

module Messages exposing (..)

import Json.Encode as JE
import Navigation
import Routing exposing (Route)


type Msg
    -- Remove this -> = FetchResult (Result Http.Error ContactList)
    = FetchSuccess JE.Value          -- Add this
    | FetchError JE.Value            -- Add this
    | Paginate Int
    | HandleSearchInput String
    | HandleFormSubmit
    | ResetSearch
    | UrlChange Navigation.Location
    | NavigateTo Route
    -- Remove this -> | FetchContactResult (Result Http.Error Contact)
    | FetchContactSuccess JE.Value   -- Add this
    | FetchContactError JE.Value     -- Add this

The previous FetchResult was receiving a Result, but now both success and error callbacks, which receive a JSON value, which need to be decoded by the update function, and we also need to update the commands calls to pass the socketUrl parameter, so let's edit the Update module:

-- web/elm/src/Update.elm

module Update exposing (..)

-- ...

update : Msg -> Model -> ( Model, Cmd Msg )
update msg model =
    case msg of
        FetchSuccess raw ->
            case JD.decodeValue contactListDecoder raw of
                Ok payload ->
                    { model | contactList = Success payload } ! []

                Err err ->
                    { model | contactList = Failure "Error while decoding contact list" } ! []

        FetchError raw ->
            { model | contactList = Failure "Error while fetching contact list" } ! []

        Paginate pageNumber ->
            model ! [ fetch model.flags.socketUrl pageNumber model.search ]

        -- ...

        HandleFormSubmit ->
            { model | contactList = Requesting } ! [ fetch model.flags.socketUrl 1 model.search ]

        ResetSearch ->
            { model | search = "" } ! [ fetch model.flags.socketUrl 1 "" ]

        -- ...

        FetchContactSuccess raw ->
            case JD.decodeValue contactDecoder raw of
                Ok payload ->
                    { model | contact = Success payload } ! []

                Err err ->
                    { model | contact = Failure "Error while decoding contact" } ! []

        FetchContactError raw ->
            { model | contact = Failure "Contact not found" } ! []

Both FetchSuccess and FetchContactSuccess branches decode the raw response received from the channel and, depending on the result, set the corresponding key value in the new model. On the other hand, FetchError and FetchContactError set a friendly error message. We are missing one thing, though; the urlUpdate function needs some minor editing as well:

-- web/elm/src/Update.elm

module Update exposing (..)

-- ...

urlUpdate : Model -> ( Model, Cmd Msg )
urlUpdate model =
    case model.route of
        HomeIndexRoute ->
            case model.contactList of
                NotRequested ->
                    model ! [ fetch model.flags.socketUrl 1 "" ]

                _ ->
                    model ! []

        ShowContactRoute id ->
            { model | contact = Requesting } ! [ fetchContact model.flags.socketUrl id ]

        _ ->
            model ! []

That is pretty much it. If we now refresh the browser and navigate through the application, everything should just work as before. Yay! If you are missing anything, please check out the branch I have prepared with all the changes for this part. I hope you have enjoyed the series and thanks for all the support.

Happy coding!