Stub controller action in Rspec controller test, but it's still executed

I would like to test if a controller action is actually called without a redirect happening in some before_filter. Since the controller action itself may do a redirect, I want to stub the action to raise a specific error (SuccessfulActionError or similar) and then check for that error as an indicator that the method was called.

So I added the following:

controller.stub!(:action).and_raise(SuccessfulActionError)

It works somehow, the exception is being raised, but the actual code in the method is still executed (e.g., if I send in the id of a non-existing record to the 'show' action, it throws a ActiveRecord::Record开发者_C百科NotFound exception).

Why is that? I want to completely stub the action as if it was implemented as

def action
  raise SuccessfulActionError
end

What am I doing wrong? It this the wrong approach?

EDIT:

Using

controller.should_receive(:action)

doesn't work either.

I overwrite the Controller in a before_all filter like this to fix default_url_options which are not picked up from the ApplicationController:

class MyController
  def default_url_options(options = {})
    { :locale => params[:locale] }
  end
end

Could that be the culprit? The specs don't work at all when I remove it unfortunately.

It's not clear to me how you can get an ActiveRecord::RecordNotFound exception after stubbing the method. The stub should replace the method call.

As an alternate approach, you might consider setting a message expectation on the action, e.g.

controller.should_receive(:show)
get :show

which should fail if a before_filter prevents the action from being called.