what did i learn today

I normally do not do a lot of view specs, but at least I want to make sure that my view renders without errors. And sometimes I really need to make sure that some link is shown or hidden depending on e.g. the role of the user or linked objects. For example, [ruby] describe "posts/show.html.haml" do context "without any comments" do it "displays no comments" do @post = Factory(:post) render rendered.should contain(I18n.t('')) end end end [/ruby] So we check the rendered result, whether it contains a specific text. But what happens if your view is rendering different yield regions? Like a body (the default region) and a sidebar. Let's use a view like this : [ruby] =show_for @post do |p| = p.attribute :title = p.attribute :content - if @post.comments = render :partial => 'comments' - else = t('') =content_for :sidebar do = link_to 'Edit', edit_post(@post) if is_allowed_to?(:edit) [/ruby] It renders the attributes using the show_for gem, and then renders inside the sidebar a link if the current user is allowed to edit it. Now I want to test what is rendered into the sidebar. To my dismay I found that neither content_for or content_for?worked at all inside rspec. And rendered does not contain the data for the other regions. So somehow I would want to get to the content for :sidebar. It appears that the different regions are actually stored inside an instance variable of the view. Once I figured that out, the rest was easy: [ruby] describe "posts/show.html.haml" do def rendered_content_for(name) view.instance_variable_get(:@_content_for)[name] end context "with enough rights" do it "displays a link to edit the post" do @post = Factory(:post) view.stub(:is_allowed_to?) { true } render rendered_content_for(:sidebar).should contain('Edit') end end context "with no rights" do it "does not display a link to edit the post" do @post = Factory(:post) view.stub(:is_allowed_to?) { false } render rendered_content_for(:sidebar).should_not contain('Edit') end end end [/ruby] Hope that this helps somebody. Or did you find a better way?

For me, it started with a tweet from Uncle Bob Martin, saying that if all your domain logic is in your model, or if you put your domain logic inside your models by default, then you are doing it wrong. I think the reasoning behind this is that we should design the domain model before relying on our database model (which is actually an implementation of that domain model). Because the best way to store something in the database is not always to best way to treat or represent or handle the data. In my Rails projects, I have to admit that in most cases the database is my domain model. Which in simple cases is correct, but in more complex cases it is not anymore. Secondly what happens is that your database model could shine through in your UI, or if it does not, your views or helpers could get very heavy. One good way to handle this is to use Presenters. Which goes into the direction of that domain logic, independent of the database. A Presenter is a class that groups all data, knows where to find it, or store it, and will match one-on-one on your presentation/view logic. Thirdly, I went to arrrrcamp, and there was Corey Haines doing his talk about Fast Rails Tests. While this title was very promising, Corey warned us that, in his own words, we would be underwhelmed. This was not entirely true, but not entirely false either :) In short what Corey said was: testing in Rails is slow because we need to drag around the framework, so why not cut out the framework where possible. So he showed an example where he extracted code from models and into separate modules, where you test them standalone. Standalone means: without requiring @spec_helper@. I did this for our modules inside @lib@ where possible, and truth be told: those specs now really fly! Awesome. But the large parts of our test-suite is our models, controllers, views, helpers where this is not possible nor wanted. I want to be able to run our complete test-suite faster, and this was not a solution for that. We scraped a few seconds of our complete time. Extracting code from our models into standalone modules will not make our total test-suite go faster either. Still it is something well worth investigating further. I don't believe in splitting up classes into modules just for the sake of making my tests faster. There has to be some logical reason (pertaining to the domain model --that is). But maybe presenters could be the way out here:

  • they group domain logic
  • while they are responsible for retrieving the correct objects from the database, these actions are just delegated to the responsible models So that sounds promising to me and a road I will investigate.
awesome rspec hints and tips

Read "Pure Rspec", a great slideshow by Jon "Lark" Larkowski. It contains some really cool rspec tips and hints, how to write your tests more readable and compact.

rails3 and rspec2 view testing

Testing the view in Rspec is just great. It allows you to test the view in total isolation. It just does the render. Make sure you set up some data that is needed, and then check if all areas are available for instance. Now for Rspec2 a lot of things have changed:

  • have_tag is no longer supported: use have_selector from webrat instead
  • response is deprecated, use rendered instead
  • the assignment has changed, write assign(:events, [stub_model(Event)]) instead of using the old assigns[:events]=[...] notation Let me show you a full example, because that always makes things clearer: [ruby wraplines="false"] require 'spec_helper' DUMMY_NUMBER="dummy number" DUMMY_MESSAGE="dummy message" describe "envelopes/index.html.haml" do before (:each) do envelope = mock_model(Envelope) envelope.should_receive(:destination).and_return(DUMMY_NUMBER) envelope.should_receive(:message).and_return(DUMMY_MESSAGE) assign(:envelopes, [envelope]) render end it ("has a h1") { rendered.should have_selector('h1')} it ("shows the correct title") { rendered.should contain(t('envelopes.list.title')) } it ("have a table-grid") { rendered.should have_selector('.grid') } it ("shows the destination") { rendered.should contain(DUMMY_NUMBER) } it ("shows the message") { rendered.should contain(DUMMY_MESSAGE) } end [/ruby]
rspec testing using macros

[ruby wraplines="false"] STATUS_EXPECTED_TRANSLATIONS = [[inp1, outp1], ... ] it "should translate received statuses" do STATUS_EXPECTED_TRANSLATIONS.each do |tr| controller.send('translate_status_to_envelope_status', tr[0]).should == tr[1] end end [/ruby] While this is short and DRY, a problem arises when one of the values in my look-up table does not behave as expected anymore. The complete test will fail, without having any clue which value caused it to fail. That is not good! But luckily, in rspec there exists the option to use macros. It is a kind of templates for tests. Using macros i could write this much cleaner, and generate my test-clauses on the fly. [ruby wraplines="false"] STATUS_EXPECTED_TRANSLATIONS.each do |tr| it "should translate #tr[0]} to success" do controller.send('translate_status_to_envelope_status', tr[0]).should == tr[1] end end [/ruby] Now if a test fails, i know which i value need to check, because i will get an appropriate error-message.

skipping slow tests in rspec2

For Test::Unit there is a module called skippy, that allows you to skip for instance slow tests. I was looking for a way to achieve the same in Rspec. I am currently developing against Rails 3.0.0.rc, and using rspec 2.0.0.beta.19 and could not find anything. So i asked on the rspec mailing list and got a great answer! It is actually present in rspec2, right out of the box, as a feature they needed for their own tests. Read the description by David himself here. He shows a lot of very good examples, even filtering tests based on ruby-version is now possible. Awesome. In short, you would do something like : [ruby] describe "something", :slow => true do it "should be blabla" end describe "something else" do it "should be a slow test", :slow => true do ... end it "should be a fast test" do .. end end [/ruby] then you can write the following to skip the slow tests : [ruby] # in spec/spec_helper.rb RSpec.configure do |c| c.filter_run_excluding :slow => true end [/ruby] and to run only the slow tests: [ruby] # in spec/spec_helper.rb RSpec.configure do |c| c.filter_run :slow => true end [/ruby] and to filter on slow tests, and run all tests if no slow tests are available: [ruby] # in spec/spec_helper.rb RSpec.configure do |c| c.filter_run :slow => true c.run_all_when_everything_filtered = true end [/ruby] Awesome! :) For the moment there is no command-line support, but in the spec/spec_helper.rb you could also check for any environment variables to do the slow or fast test-set. What i personally needed, was something that would either run all tests, or run all tests excluding a few (i have a test-suite for an sms-gateway and do not want to send sms-es all the time). So i handled that like this: [ruby] # in spec/spec_helper.rb RSpec.configure do |config| if ENV['SEND_REAL_SMS'] != "true" config.filter_run_excluding :send_sms => true end end [/ruby]

missing 'a' with rspec and cucumber

I am developing Ruby on Rails on Windows platforms mostly. But using Rspec and cucumber on windows has a very strange side-effect: all a's are missing as can be seen from the following screenshot: rspec-without-a-smallerLuckily, after some very extensive googling, i found a single blogpost with a fix! Apparently it has something to do with UTF-8 encoding, and the simple solution is that you need to change the encoding of the current command prompt. This can be achieved via a simple call before you start: [sourcecode] chcp 1252 [/sourcecode] The aforementioned post then proposes to adjust the cucumber.bat to not have to type this every time. This is all good for cucumber, but not for rspec, and anyhow, every new update of the cucumber i would need to apply this fix again. I was thinking that it might be possible to set the default codepage, which is 850 on my machine, to 1252 permanently. As this blogpost mentions, there are two different ways to achieve the wanted result.

Change the codepage system-wide

This can be done in the registry. [sourcecode] [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Nls\CodePage] OEMCP=1252 [/sourcecode] But one commenter notes this is not without risk.

Only for command prompt

An alternative way is to put make sure that each time a console is opened (cmd.exe) the codepage is set automatically. [sourcecode] [HKEY_LOCAL_MACHINE\Software\Microsoft\Command Processor] Autorun=chcp 1252 [/sourcecode] This will only work for console windows in which you run cmd.exe, which is just what i needed.