View specs are great. While I always keep in mind they are a unit test and do not test the full integration of a feature, I find that they are fantastic for asserting around all the edge cases of a feature - things like the expected validation messages showing up, or buttons appearing or disappearing based on different scenarios. I used to use system tests for this kind of thing, and find that these get really slow and fragile over time, while view specs stay blazing fast.

Often, I’ll break a template up into smaller pieces. Rails has a mechanism for this called partials. I’ll often break up things that are either their own “thing” (like a “card” display for an object, or “share links” for an object), or things that have some logic wrapped around them that I want to keep contained (for example, a button with a permission check wrapped around it).

When I’m testing such templates with a view spec, I will often have a view spec for the main template, and then another view spec for any more complex partials. In this scenario, I want to assert in the main template that the partial was rendered, but I don’t mind too much what is rendered, since I have another view spec specifically for that partial that checks that.

In this case, I’ve found that RSpec spies are a great way to do this. Spies are set up ahead of the test run, record interactions with themselves, and then allow the calls that they received to be asserted against. RSpec spies also have a handy method called and_call_original. This records the call, but then invokes the original method. I can use this to record that a particular call to render was received, but then still actually render the partial.

Here’s an example:

require "spec_helper"

RSpec.describe "widgets/show.html.erb", type: :view do
  it "renders the 'edit_button' partial" do
    allow(view).to receive(:render).and_call_original

    expect(view).to have_received(:render).with("widgets/edit_button", widget: widget)

This test catches if the correct partial is not rendered, or if unexpected locals are passed to the partial, but doesn’t require me to assert in this spec exactly what the partial might render - I can put that in the partial spec.

I’ve found this is quite a handy technique and helps to make it quick and easy to test templates, since it allows the test to be super focussed on what the current template under test is doing without worrying about any conditions the parent template might otherwise need to take into account. It also allows for some nice dependency injection into the partial spec, since the locals can be adjusted in the spec (for example, to pass a different ‘user’ local, or a feature flag stub or something like that). Changing these settings just for one partial in a wider template would not otherwise be possible without splitting up the template into partials, and the template spec into partial specs.

This technique is also a candidate for refactoring into a custom matcher for view specs. I won’t cover that in this post, but since allow(view).to receive(:render).and_call_original doesn’t actually effect how templates are rendered, you could quite easily run this before all view specs, and then support a matcher like expect(view).to have_rendered_template() - which is very similar to what the rails-controller-testing gem does - but in a view spec rather than a controller spec (though I suspect this gem probably would work with a view spec since a view spec defines an anonymous controller under the hood).