let's build an elm app

November 9, 2018

in the beginning...

there was a browser, and it was good.

there were also machines, science, and the lamba calculus. they were also good.

together, they inspired elm lang. it is good.

elm is for making browser applications. it has a great guide, too. although the guide is thorough, it is unabashedly terse, much like elm itself. i would argue that the elm guide's conciseness is a feature™. however, when it comes down to actually writing code, those of us who have forgetten our functional roots, or those of us who were never introduced, have a hard time knowing how to model our programs.

so, let's write some code. below we will make a small app, from scratch, with milestones on github for those who want to code along. forewarning, this will not cover many of the topics in the guide. i will try and drop helpful breadcrumbs into the guide as we progress.

baby's first steps

let's make an app about bikes. 🚲

the purpose of this website will be to convince everyone to go biking with me, and give them an interactive scroll experience as they click around an in-app survey.

first, install node & elm.

next, run:

$ npx create-elm-app up-and-up && \
  cd up-and-up && \
  elm-app start
You can now view up-and-up in the browser.

  Local:            http://localhost:3000/
  On Your Network:

sweet. our first app. let's take a peek in the browser:

app init

git milestone 1

there's really not much to see yet. i removed a bit of cruft from the default boilerplate in git milestone 2, which is where we actually begin.

bootstrap the app

like so many languages, your elm program begins with a main function. it's just like old times!

main =
        { view = view
        , init = \_ -> init
        , update = update
        , subscriptions = always Sub.none

the Browser.element function is critical to our app. it takes a record and wires up other functions that tell our app:

  1. what to render (view),
  2. what state to start with (init),
  3. how to react to events/side-effects (update), and
  4. how to listen to side-effects (subscriptions)

familiar with react & redux? here's what these fields loosely conceptually map into:

{ view = view -- a functional stateless component, at the root of your app
, init = \_ -> init -- getInitialState()
, update = update -- rootReducer
, subscriptions = always Sub.none -- redux-observable, redux-saga

not so bad right? the full signature is actually:

main : Program () Model Msg
main = Browser.element { ... }

the signature of main takes no args and returns a... Program () Model Msg? if you are thinking "WTF is a Program () Model Msg", don't worry about it--seriously. it will come naturally later.

let's make some DOM!

we just saw the view function used above. very soon, we will want to add some bicycling content to the page. before we do, let's break down the current view function. we began with:

view : Model -> Html Msg
view model =
    div []
        [ img [ src "/logo.svg" ] []
        , h1 [] [ text "Your Elm App is working!" ]

looking at the first line, we see that view is a function that takes as input a Model instance and produces a Html Msg.

next, let's break the body down bit by bit:

view model = ...
-- fn_name arg = <body>
-- div is a fn from the `Html` package with the signature:
-- fn_name (List Html.Attribute) (List (Html Msg))
-- which outputs a Html Msg
div [] [...]
-- children of the div
-- img, h1, & text are functions from `Html` module
  [ img [ src "/logo.svg" ] [] -- src is a function from the `Html.Attributes` module
  , h1 [] [ text "Your Elm App is working!" ]

in the end, view is a just function that creates a Html Cmd instance. for now, you can think of this as just Html and ignore the Cmd part if you are unfamiliar. these Html instances will eventually flush out to DOM nodes, if you hadn't already guessed. view calls a bunch of other functions internally as well with Lists of attributes and Lists of children, where each of those also produce Html. if you've done any react development, you may notice how similar this signature is to React.createClass(type, props, children).

let's get a bike image on screen, and some info about the bike club.

view model =
    div []
      [ header []
          [ img [ src "/bike.gif"] []
          , h1 [] [ text "Up n' Up™" ]
          , h4 [] [ text "let's go biking!" ]
      , div [ class "wut info" ]
          [ h1 [] [ text "what is it." ]
          , p [] [ text "you. me. us.  biking!" ]
          , p [] [ text "let's ride bikes at lunch." ]
      , div [ class "when info" ]
          [ h1 [] [ text "when is it." ]
          , p [] [ text "wednesdays! noon. fair weather only." ]

let's add some image and css assets, and see what we get:

"we didn't need elm to do that" you may be saying. we sure didn't, but isn't it beautiful? who wants to write a loosey-goosey DSL (HTML) over strongly typed, terse functions instead? git milestone 3

loadin datas

as discussed earlier, i want to make the page interactive for my viewers. the goal is to ask the user a question, and provide a fixed set of responses for them to respond to. upon clicking a response, the user will be scrolled to the next question, and so on, until finally they end up agreeing to go biking. we will make a data structure to represent this basic graph traversal exercise. it's worth noting that our question set will have cycles in it. we want to trick the user into thinking that he or she still has free will in this decision. bikes or bust!

here's a snapshot of the question graph:

    "key": "READY",
    "text": "ready to ride?",
    "options": [
      { "text": "i'm in!", "to": "JOIN_BIKE_CLUB" },
      { "text": "no. i'm not interested.", "to": "HATE_BIKES" }
    "key": "HATE_BIKES",
    "text": "what--do you hate bikes?",
    "options": [
      { "text": "ya.  i hate 'em real bad", "to": "YOU_DONT_HATE_BIKES" },
      { "text": "no.", "to": "WAS_IT_SOMETHING_I_SAID" }
    "key": "JOIN_BIKE_CLUB",
    "text": "nice! let's ride!",
    "options": [
      { "text": "what's next?", "to": "GET_IN_TOUCH" }

i'll create a new elm module to define types to marshall these values into:

-- Node.elm
module Node exposing (Edge, Node)

import List

type alias Edge =
    { text : String
    , to : String

type alias Node =
    { key : String
    , text : String
    , options : List Edge

now, how do we actually load our data into Elm? because the json is static, we can use elm's flags feature to provide this data on init. but before we get to using flags to load to data into elm, we need to figure out how to load the asset into the browser at all.

each elm application's true entrypoint is actually a js file. in this application we are using create-elm-app, which uses webpack to bundle and load assets, available to us for free. in this context, we can simply import <json-file>, and pass that file directly into elm.

import "./main.css";
import { Elm } from "./Main.elm";
import nodes from "./nodes.json";

  node: document.getElementById("root"),
  flags: {
    // `nodes` is parsed into a javascript object by webpack's
    // json-loader.  it is _not_ a JSON string

neat. now that our data is available and being passed, we actually need to do something with it, elm side.

first off, we need to define types for flags. it needs to have the exact same shape as what is passed to Elm.Main.init({ flags: FLAGS }):

type alias Flags =
    { nodes : Array Node -- `List Node` would work, too

when we get the flags data, what will we do with it? the elm architecture has a recipe for this. we need to update our Model to capture and persist the inbound data:

type alias Model =
    { nodes : Array Node

as discussed earlier, the init function is responsible for mapping flags to our initial state. let's update the default init function:

init : Flags -> ( Model, Cmd Msg )
init flags =
    ( { nodes = flags.nodes
    , Cmd.none

finally, let's update our main to reflect this new workflow:

main : Program Flags Model Msg -- `Flags` was previously `()`
main =
        { view = view
        , init = \flags -> init flags
        , update = update
        , subscriptions = always Sub.none

not too bad, right? it feels like a bit much at first, but we did a bit of extra legwork to make sure that the js -> elm interop went swimmingly. a takeaway that is not immediately apparent in the above code is that when flags are passed, they are typechecked at runtime as they are coerced into elm types!

at this point, in our app we can see that our model has been successfully updated via the elm debugger:

nodes loaded

see git milestone 4.

render the nodes

model acquired! let's render the nodes onto the page:

-- refactor our introductory content into its own function
render_intro =
    [ header []
        [ img [ src "/bike.gif" ] []
        , h1 [] [ text "Up n' Up™" ]
        , h4 [] [ text "let's go biking!" ]
    , ... -- truncated for brevity

render_node node =
        { options, text } =
    div [ class "node__container" ]
        [ div
            [ class "node__content" ]
                [ [ h1 [] [ Html.text text ] ]
                , map
                    (\link ->
                        div []
                            [ button [ onClick NoOp ] [ Html.text link.text ]

render_nodes nodes =
        fragments =
            Array.map render_node nodes
        Array.toList fragments

view model =
    div [] <|
            [ render_intro
            , render_nodes model.nodes

what a simple view! the primary view function renders two main segments: the introduction section and the nodes section. render_nodes calls render_node for each node. it's rather boring. with a little css spritzed in, we have a crude rendering implementation in place. see see git milestone 5.

forgive the fact that nothing is particularly pretty by this point. let's instead take a peek at that NoOp used above.

get interactive

when we click a button, it should navigate the edge to the associated node. but all we have is a NoOp! let's swap it with a ScrollToPane String.

type Msg
  = NoOp
  | ScrollToPane String -- new Msg type!

scrollToPane id = -- new function! returns a Cmd msg
    Dom.getElement id
        |> Task.andThen (\el -> Dom.setViewport el.element.x el.element.y)
        |> Task.onError (\_ -> Task.succeed ()) -- hack. swallow DOM errors for now
        |> Task.perform (\_ -> NoOp)

update : Msg -> Model -> ( Model, Cmd Msg )
update msg model =
    case msg of
        NoOp ->
            ( model, Cmd.none )

        ScrollToPane id -> -- new case!
            -- upon request, kick off the managed scroll side effect
            ( model, scrollToPane id )


render_node node =
    -- new attribute! we're adding an id now to each node root
    div [ id key, class "node__container" ]
       ... -- and let's swap that `NoOp` with a `Msg` of `ScrollToPane String`
       (\link ->
          [ button [ onClick <| ScrollToPane link.to ] ...

now the buttons actually work!

  1. onClick, we emit a ScrollToPane String message.
    1. remember, the view function produces a Html Msg--Html that can emit messages! Browser.element wires those messages to our update handler.
  2. the update function handles the message and triggers elm to go off and work on the command generated by scrollToPane.
  3. scrollToPane queries the element id we passed in our message above and scrolls to it.
  4. we NoOp post-scroll. clicking a navigation option only generates side-effect.

no sweat. git milestone 6 complete.



there's so much more to do and cover! in fact, in my alternate implementation (src), i dig into using elm ports and subscriptions, which are actually fairly straightforward js to elm pubsub mechanisms. i also do a bit more styling.

regardless, let's summarize. we successfully:

  1. created an elm app using create-elm-app
  2. studied the elm bootstrap sequeunce (Elm.Main.init -> main)
  3. understood how elm uses pure functions to create Html/Dom nodes
  4. learned how attributes and children can be wired in, using additional functions from the Html[.*] modules
  5. grazed over the elm architecture, which we drew direct parallels to react/redux
  6. updated our applications state model and messages
  7. bound our view layer to a click handler that generated messages that could be reduces
  8. fired off a managed side-effect for scrolling the browser, and got a first taste of elm Tasks, which are akin to lazy promises.

questions? edits? send them my way here.