Best Practices for Testing Phoenix (Functional) Components

Are there any recommended patterns or best practices for testing Phoenix Components? In particular, how thorough should a test be, and what specifically should be verified?

Since Phoenix components are a newer feature, one might consider looking to existing Phoenix approaches to testing rendered HTML. In that case, the standard way seems to follow the form assert my_html =~ a_string (see, for example, the code snippets at Testing Controllers — Phoenix v1.6.6 or Phoenix.ConnTest — Phoenix v1.6.6, or of course the tests generated by mix phx.gen.html).

This approach may be sufficient when the test subject is a controller, but in the case of view component that will be reused across many contexts, I wonder if it is thorough enough.

Example component

Consider, for example, the table function component outlined in the docs (Phoenix.LiveView.Helpers — Phoenix LiveView v0.17.5):

def table(assigns) do
  ~H"""
  <table>
    <th>
      <%= for col <- @col do %>
        <td><%= col.label %></td>
      <% end >
    </th>
    <%= for row <- @rows do %>
      <tr>
        <%= for col <- @col do %>
          <td><%= render_slot(col, row) %></td>
        <% end %>
      </tr>
    <% end %>
  </table>
  """
end

To test it, would I want to:

  • assert that certain strings appear in the output, a la assert rendered_to_string(my_table_heex) =~ my_col.label ?
  • verify the HTML structure itself using something like Floki?
  • …or something else altogether?

Example test

I might test such a table component like so:

defmodule TableComponentTest do
  import Phoenix.LiveView.Helpers
  import Phoenix.LiveViewTest

  # Example test
  test "table" do
    # the ~H sigil expects an `assigns` variable to be defined
    assigns = []
    item = %{name: "an item", category: "a category"}
    col1 = %{label: "a column"}
    col2 = %{label: "another column"}

    table_str =
      rendered_to_string(~H"""
      <TableComponent.table rows={[item]}>
        <:col let={item} label={col1.label}>
          <%= item.name %>
        </:col>

        <:col let={item} label={col2.label}>
          <%= item.category %>
        </:col>
      </TableComponent.table>
      """)

    # EXAMPLE ASSERTIONS
    # Option 1
    assert table_str
           |> Floki.parse_fragment!()
           |> Floki.find("th td")
           |> Floki.raw_html() == "<td>#{col1.label}</td><td>#{col2.label}</td>"
    # etc.

    # Option 2
    for fragment <- [col1.label, col2.label, item.name, item.category] do
      assert table_str =~ fragment
    end
  end
end

Option 1 feels thorough, but I don’t typically like checking return value internals if I can help it. On the other hand, Option 2 feels too lax and doesn’t give me enough confidence that I can use the component in practice. I feel like I need a middle way.

Outside of the Elixir ecosystem, I’ve found the ViewComponent (Rails) testing guide helpful.

Do feel free to share any thoughts, ideas, and experience! If there is an officially recommended approach, I would welcome it; if not, perhaps discussions like this could help lead to one.

2 Likes

In both cases, matching HTML output against the inputs with == can cause flaky tests if you’re using randomized data via faker or similar: an assertion like Floki.raw_html() == "<td>#{col1.label}</td><td>#{col2.label}</td>" will fail if col1.label or col2.label contain characters that are escaped in HTML. For instance, sometimes generating a last name with Faker will produce O'Brien which escapes to O&apos;Brien :frowning:

Ah good point – I am randomizing test data and would likely run into this at some point.

So how should the expected value in an assertion be formatted? As an HTML-escaped string? Or something else?

Would Wallaby be a good fit for unit testing Phoenix Components?

A different point of view, drawing on this talk by Ian Cooper.

What if function components are the wrong level altogether to test? Maybe the correct granularity is where the component is used in context, and it is not necessary to test the function component in isolation.

With the table component, that would mean instead of testing the table component directly, one would test the live view or controller that uses the component. If the live view or controller renders the expected elements, then the behavior has been verified. Whether a component is used or not is an implementation detail. The developer should be free to extract or inline components as desired without worrying about breaking tests.

With LiveViewTest you can very precisely target html elements using advanced selectors and check their content. You don’t need Floki nor Wallaby I guess. You probably need those only for static pages.

See element selector here: Phoenix.LiveViewTest — Phoenix LiveView v0.17.5

1 Like

Thank you for the suggestion. Looking at the docs and source for Phoenix.LiveViewTest.element/3, it seems like the function requires a %Phoenix.LiveViewTest.View{} to be passed. The same appears true for most helper functions in LiveViewTest.

Given that, is it possible to use these helper functions when testing a functional component in isolation? That is, is it possible to create and pass a %View{} when not testing an actual Live View? If so, then the toolkit provided by LiveViewTest should be sufficient for navigating and asserting on functional components.

The section on testing function components recommends two functions: LiveViewTest.render_component/3 and LiveViewTest.rendered_to_string/1. The examples provided don’t appear to do more than simply asserting string equality.

With Conn Tests (for “dead views,” as it were), we can generate a %Conn{} struct for testing using ConnTest.build_conn/0. Is the same possible for a %View{} in a function component test?

1 Like

I was about to ask exactly this question.
How do I leverage the more powerful test-helpers like has_element? with a function component?