If you are coming to the Phoenix Framework from another framework such as Ruby on Rails, Python’s Django, or PHP’s Laravel, you will notice many similarities between your current framework and Phoenix. Many of these commonalities are very strong, such as breaking up a project into Models, Views, and Controllers (i.e. MVC), while others will have you scratching your head asking if the developers are even from the same planet. There will still be other areas of Phoenix which will appear very familiar, but only as long as you don’t look below the surface. Phoenix Templates are an example of this latter group.

As in most web frameworks, Phoenix templates are files containing a mixture of static markup and the framework’s underlying code; in the case of Phoenix, the underlying code is Elixir. These templates can take the form of HTML, XML, JSON, plain text, or any other textual file type. Where Phoenix templates differ, however, is that they are actually compiled into functions. While this alone would be enough to surpass the performance of other frameworks, Phoenix has yet another trick up its sleeve:

After compilation, templates are functions. Since Phoenix builds templates using linked lists rather than string concatenation the way many imperative languages do, one of the traditional bottlenecks of many web frameworks goes away. Phoenix doesn’t have to make huge copies of giant strings.

– Chris McCord, Programming Phoenix 1.4

We’ll see more about how this works shortly, but for now it’s enough to know that Phoenix templates are performant and memory friendly.

Template Location and Naming

Phoenix templates are kept under the “templates” directory of your project’s web directory, and are grouped by the View to which they are associated. For example, if you have a PageView.ex in your app named “Sprawl”, the templates would be located in lib/sprawl_web/templates/page/. Other frameworks might break this down by Controller rather than View, but in Phoenix, Templates and Views are completely different.

Templates follow this naming convention: <action>.<content_type>.eex. “action” usually, but not always, corresponds to the Controller action, while “content_type” is the requested content type of the data: HTML, JSON, XML, etc. Lastly, .eex refers to Elixir’s EEx library which handles transforming the template files into their final form. A typical HTML request hitting the index action of a PageController would result in a page being generated from the templates/page/index.html.eex file.

Generating Output

Before turning our attention to building templates, the syntax, and working with variables and helpers, it’s useful to understand what happens between the initial call to the controller action and producing the final output. You’ve likely deduced the high-level flow. It looks something like this:

web_request
|> router
|> controller
|> view
|> template

At this height, however, you miss certain details such as where the layout is set, how the view is determined, and why Phoenix templates are so amazingly fast. We can answer those first two questions by zooming in on a typical controller:

# lib/sprawl_app/controllers/page_controller.ex

defmodule SprawlWeb.PageController do
  use SprawlWeb, :controller

  def index(conn, _params) do
    render(conn, :index, name: "Molly")
  end
end

In this example controller, notice that the first thing the controller does is use the controller function from the SprawlWeb module. This module is generated at the time of project creation and can be found at lib/sprawl_web.ex. If we look at that file, we see we’re entering the world of macros (the quote macro gives it away). Don’t worry though, we’re only going deep enough to understand the flow:

defmodule SprawlWeb do
  def controller do
    quote do
      use Phoenix.Controller, namespace: SprawlWeb

      import Plug.Conn
      import SprawlWeb.Gettext
      alias SprawlWeb.Router.Helpers, as: Routes
    end
  end

  ...
end

Like our PageController above, the first thing the controller function does is use another module, Phoenix.Controller. Let’s take a peek at that:

defmodule Phoenix.Controller do

  ...

  defmacro __using__(opts) do
    quote bind_quoted: [opts: opts] do
      import Phoenix.Controller

      # TODO v2: No longer automatically import dependencies
      import Plug.Conn

      use Phoenix.Controller.Pipeline, opts

      plug :put_new_layout, {Phoenix.Controller.__layout__(__MODULE__, opts), :app}
      plug :put_new_view, Phoenix.Controller.__view__(__MODULE__)
    end

  ...

end

This module is…less straightforward, but the important part – at least as far as we’re concerned – is the last two lines. The first of those two lines sets the initial layout to be app (i.e. templates/layouts/app.html.eex. The second line sets the view to be one matching the calling Controller’s name. Both lines use the plug/2 macro, which tells us they’re just setting a field in the %Plug.Conn{} struct. It also tells us that we’re free to change the layout and view in the actions we define.

Once our PageController has use-ed the Phoenix.Controller module, it’s finally able to render the index action. There are two things that might surprise you when it comes to what Phoenix does when rendering a template: 1) it stores the entire HTML document in the :resp_body field of the %Plug.Conn{} struct; 2) it stores that HTML document as an IO List, not as a String. An IO List is just “a list of things suitable for input/output operations.” We can get a glimpse of what that looks like by rendering a template with a View.

Here’s an example template:

# lib/sprawl_app/templates/page/index.html.eex

<h1>Hi, <%= @name %></h1>

If we render that in the console, it will look something like this:

iex > Phoenix.View.render_to_iodata(SprawlWeb.PageView, "index.html", conn: %Plug.Conn{}, name: "Wintermute")

[[["" | "<h1>Hi, "] | "Wintermute"] | "</h1>\n"]

The result is just a nested List (technically an improper list) which Phoenix sends directly to the I/O stream. Because Phoenix responds in this manner, there is no need to concatenate the file into one large string, consuming needless memory. On top of that, handling I/O this way allows the BEAM to use the same memory address for duplicate strings, further reducing the server’s memory needs:

…the function that Phoenix compiles to render that template will use the same strings in memory every time, simply adding the dynamic parts as needed to the IO list. This is, in effect, a kind of view caching.

Elixir and IO Lists, Part 2: IO Lists in Phoenix

That’s a bit of “why” Phoenix templates can be so much faster and performant than other frameworks. Now it’s time to look at how to build them.

EEx

By default, Phoenix templates use Elixir’s built in templating engine, EEx. EEx stands for “Embedded Elixir” and as you might expect , “it allows you to embed Elixir Code inside a string.” If you are familiar with Ruby’s ERB templating system, you’ll already be familiar with EEx’s syntax. An example HTML file using the EEx templating system looks like this:

<h1>Users</h1>

<%= link "New User", to: new_user_path(@conn, :index) %>

<ul>
  <%= for user <- @users do %>
    <li><%= user.name %> (<%= user.age %>)</li>
  <% end %>
</ul>

There are four different EEx tags, two of which are used above.

<% %>: Inline

Use the inline tag when you need to perform some bit of logic without displaying the results in the output. Examples might be calculating a result or setting a variable. More often, however, you’ll use the tag to end blocks of code or the like.

# <% %> example

<% fullname = user.first_name <> " " <> user.last_name %>

Hi, <%= fullname %>!

<%= if true do %>
  Greetings, Program!
<% end %> # <- end

<%= %>: Output capture

Unlike the inline tag, the output capture tag displays the output of the embedded elixir code, doing so even across multiple lines. Unlike Ruby’s ERB syntax, you will use output capture tags with if, case, and other logic structures, because these elements are functions which have return values.

# <%= %> example

<%= if true do %>
  this line will be output
<% else %>
  this line never will
<% end %>

<%= case response do %>
  <%= {:ok, response} -> %>
    something worked!

  <%= {:error, :whoops} -> %>
    that didn't work.

  <%= _ -> %>
    I have no idea what happened
<% end %>

<%# %>: Comment

The comment tag is used – to no one’s great surprise – to add comments. More often, however, it’s used to temporarily comment out code within the template. Unlike HTML comments <!-- -->, EEx comments are never output into the final product.

# <%# %> example

<%# if true do %>
  this will always be output
<%# else %>
  So will this as long as the "if" is commented out
<%# end %>

<%% %>: Quotation

Quotation tags are used when you need to display EEx syntax in the output without actually processing the code within the tag. An example might be to show an example of Elixir’s EEx template on a webpage. More likely, however, you’ll use HTML’s built in character entities.

# <%% %> example

<p>Check out this whizbang elixir code!</p>

<code>
<%% String.reverse("!edoc rixile gnabzihw") %>
</code>

Variables

Templates wouldn’t be very useful if there wasn’t a way to get data from Models and Controllers to the View layer. To that end, we can “assign” values to variables when we render templates in our controllers, views, or even other templates.

# Variable assignment example

# lib/sprawl_web/controllers/page_controller.ex
defmodule SprawlWeb.PageController do
  use SprawlWeb, :controller

  def index(conn, _params) do
    render(conn, :index, name: "Wintermute")
  end
end

# lib/sprawl_web/templates/page/index.html.eex

<h1>Hi, <%= @name %></h1>

In this example we are “assigning” the value “Wintermute” to the key name. We are then able to access the name variable in our template by prepending the @ character to it. In this case, name is just a string, but we could just as easily assign a List, Map, or a more complex data structure. Furthermore, we are only assigning a single variable in the example, but multiple values can be set here as well.

Helper functions (Phoenix.HTML)

HTML isn’t a complex language. There’s no logic, the syntax is simple, and a person can learn the basics in less than an hour. Why then does Phoenix provide a handful of tools deemed “Helper functions” to write HTML and aid in the creation of web pages? There are three primary reasons:

Helper Functions Simplify Tag Creation

Tags such as <a> and <img> are used frequently enough in webpages that not only is it simpler to use a helper function, but their usage makes reading the HTML easier. Judge for yourself, which do you find easier to read?

# Interpolated
<a href="<%= page_path(@conn) %>">Home</a>
<img src="<%= static_path("logo.png") %>">

# Helper function
<%= link "home", to: page_path(@conn) %>
<%= img_tag static_path("logo.png") %>

Not only do these helper functions make the code easier to read, but the helper functions return an IO List which we’ve already seen is more memory friendly. It makes sense for link/2 and img_tag/2 to be helper functions because it mitigates string interpolation and the likelihood of mistyping the syntax, but why have a function like button/2?

Helper Functions Automate Security

Both button/2 and link/2 accept a method: attribute to define what type of HTTP call should be made when the user clicks on the respective tag. If that method is anything other than get, these functions include a data-csrf attribute and token in the HTML tag, which hinder CSRF attacks.

Furthermore, when you use any of Phoenix’s helper functions, they sanitize any attributes and content which are set programmatically. This is especially useful when presenting user-provided data.

# Potentially evil JavaScript
user_data = ~s["><script>alert("Evil script!")</script>]

# l
Phoenix.HTML.Tag.content_tag(:p, user_data)

# outputs to

{:safe,
 [
   60,
   "p",
   [],
   62,
   [
     [
       [
         [
           [[[[[] | "&quot;"] | "&gt;"] | "&lt;"], "script" | "&gt;"],
           "alert(" |
           "&quot;"
         ],
         "Evil script!" |
         "&quot;"
       ],
       ")" |
       "&lt;"
     ],
     "/script" |
     "&gt;"
   ],
   60,
   47,
   "p",
   62
 ]}

As you can make out, the potentially damaging script has been neutralized.

Helper Functions Allow Programmatic HTML Creation

The last reason to use Phoenix’s helper functions is to give you the ability to programmatically create HTML, such as from View functions. In this case, you want to make sure the generated HTML is safe. Phoenix’s helper functions ensure that it will be.

defmodule PageView do
  def component do
    content_tag(:div,
      content_tag(:p, "The sky above the port was the color of television, tuned to a dead channel."),
    class: "card")
  end
end

The output of the above results in the following:

{:safe,
 [
   60,
   "div",
   [[32, "class", 61, 34, "card", 34]],
   62,
   [60, "p", [], 62,
    "The sky above the port was the color of television, tuned to a dead channel.",
    60, 47, "p", 62],
   60,
   47,
   "div",
   62
 ]}

Although this is a possible use case, it’s not commonly used, nor is it generally recommended. Instead, it’s usually better to render smaller templates, which I’ll discuss in a future article.

Summary

Every web framework has its own ideas about how templates should behave, and Phoenix is no different. Where Phoenix differs is in how it outputs the final results. Rather than relying on concatenating a myriad of strings together into a final form before sending it off, Phoenix relies on Erlang’s IO Lists to both increase its responsiveness and limit memory usage.

By default Phoenix uses Elixir’s built in EEx (Embedded Elixir) library to handle template compilation. With only a handful of tags and helper functions to learn, EEx makes working with HTML and other textual file types eminently approachable to both practiced developers and novices alike.

In the next article, we’ll look more closely at how to render smaller templates within the larger context, the best practices in doing so, and also how to work with layouts.

Further Reading