render_to_string :json (or :xml or :update) before render results in DoubleRenderError

Automatic cleanup of spam.

Although abstract_controller/rendering.rb:91 would lead you to believe that response_body is only set when you call render (and not when you call render_to_string), that isn't the case in Rails 3.0.3. If you try an experiment along the following lines, you'll see that response_body has been written:

def index
  test = render_to_string(:template => "main/index")
  Rails.logger.info response_body
  # ...
end

Because response_body is not nil, action_controller/metal/rendering.rb:16 raises AbstractController::DoubleRenderError. You'd think that you can refactor your example like this:

class TestController < ApplicationController
  def test
    json = render_to_string(:json => "[]")
    response_body = nil
    render :text => json
  end
end

but there's a catch. There's a second response_body accessor! It's in abstract_controller/base.rb:14. So what you'd actually write is:

class TestController < ApplicationController
  def test
    json = render_to_string(:json => "[]")
    @_response_body = nil
    render :text => json
  end
end

That should serve as a workaround for now.

This is an edge case. You will run into this error only if you are using

render_to_string(:json => '')
render_to_string(:xml => '')
render_to_string(:js => '')

That is because of following code

    add :json do |json, options|
      json = json.to_json(options) unless json.kind_of?(String)
      json = "#{options[:callback]}(#{json})" unless options[:callback].blank?
      self.content_type ||= Mime::JSON
      self.response_body  = json
    end

    add :js do |js, options|
      self.content_type ||= Mime::JS
      self.response_body  = js.respond_to?(:to_js) ? js.to_js(options) : js
    end

    add :xml do |xml, options|
      self.content_type ||= Mime::XML
      self.response_body  = xml.respond_to?(:to_xml) ? xml.to_xml(options) : xml
    end

I would say that render_to_string should always ensure that response_body is set to nil. I have attached a patch and a test.

Applied, thanks Neeraj!