If somebody asked me if I used transactions much in MySQL, I would probably have said no… however, now that I don’t have them I realise just how much I used them. Things that just work when using ActiveRecord such as creating a record and ensuring that an associated record gets created is sooo much harder.

MongoDB has “Atomic Operations” which means if you have an embedded relationship, all is good and the database will be rolled back if the entire document does not save. As soon as you have two top level collections and need to ensure that a record is created in both, things get a little hairy.

There seems to be a great deal of “if you need transactions just use a RDBMS” talk but what if you need a flexible schema and transactions, the last thing you want to manage is some kind of EAV system like Magento.

I’ve created a thread on the Mongo User forum with a typical example so feel free to join in the discussion. I’d love to hear experienced Mongo user’s views on how to overcome these kinds of situations and what kind of applications people are running with MongoDB.

Dennis Reimann has posted a nice alternative here using Capybara… thanks Dennis.

IQ::Color is a really simple Gem for converting colour values to and from CSS(3) notation, this is used by a couple of our larger Gems but I can see it being of use on it’s own. A simple example:

color1 = IQ::Color.RGB.new(127, 127, 255)
color1.to_css #=> "#7f7fff"
 
color2 = IQ::Color.from_css('rgba(127,127,255,0.4)')
color2.red    #=> 127
color2.green  #=> 127
color2.blue   #=> 255
color2.alpha  #=> 0.4
color2.to_css #=> "rgba(127,127,255,0.4)"

More examples can be found on GitHub.

If anyone else gets the same problem, just launch Activity Monitor which lives in the /Applications/Utilities folder, click the CPU tab if not already selected, select processes with a process name of “ruby”, click “Quit Process” and then select “Force Quit”. Just selecting “Quit” wouldn’t work for me hence the “Force Quit”.

I hope this saves someone else the head-scratching as to why their Mac is running hot.

In the past I have used FakeWeb, however it becomes a little hairy when you need to stub a response based on request body. I came across a couple of alternatives, firstly WebMock which looks promising but then Artifice from the mighty Yehuda Katz caught my eye…

Artifice lets you “replace the Net::HTTP subsystem of Ruby with an equivalent that routes all requests to a Rack application”. I like the simplicity of this solution as you can in essence use a Rack application to replace the responses of the service you are testing.

An Example

First we create a simple Rack application to stand in for the server we are interacting with.

app = proc do |env|
  [200, { "Content-Type"  => "text/html" },
    ["Hello world: #{env.inspect}"]
  ]
end

Then we simply use Artifice’s “activate_with” method to wrap any requests.

Artifice.activate_with(app) do
  response = Net::HTTP.start("google.com") do |http|
    http.post("/the_url", "foo=bar")
  end
  puts response.body
end

This allows for a Rack app to be used as a stand in for a complete API, it could be a Sinatra app for example allowing for easy route handling. We could go so far as to have a series of Rack apps that can be used as stand-ins for common API’s.

I have lately been using Pickle when writing Cucumber features, however I have come to the conclusion that this is a bad idea. The reason being that when using Pickle, you create entries directly, whereas the whole point of Cucumber is that it is for high level integration testing.

What I do now, is to create any entries by filling out and submitting the relevant forms with a step definition, for example I may have the following:

Given an admin has created the following products
  | Name    | Variants                                  | Featured |
  | T-Shirt | Small: 10.99, Medium: 12.99, Large: 14.99 | Yes      |
  | Keyring | Default: 2.99                             | Yes      |

I would then write a step definition for this that would log in as the admin user, break this table appart and fill in the relevant forms.

I would go so far as to say that using factories at all in features is a bad idea and instead everything should happen via the user interface for better coverage. For any data that is known to exist when the app is deployed via ‘rake db:seed’, this can be loaded in the ‘env.rb’ file e.g.

load 'seeds.rb'

Update 11/2/2010: Sometimes this is simply not practical due to slowdown which is a shame, as noted by Amos in the comments.

Devise uses Warden, which is an authentication solution build on Rack making it extremely flexible and usable across multiple frameworks e.g. Rails/Sinatra. Also, Devise is extremely modular meaning to can easily write custom “strategies” for specific behaviour.

I have used Clearance in the past which is great if you want an engine that will just work. Devise however is by far the most flexible and extensible solution that I have come across with the same ease of use as Clearance. The only thing that you don’t get with devise that you do with Clearance is the signup stage, however as this is normally custom on a per-app basis I can live with this.

One more thing to note is that Devise lets you have multiple auth systems in play e.g. one for users and one for admins.

Using Memoization as an example:

class MyClass
  def lazy_initialized_value
    @lazy_initialized_value ||= Expensive.request
  end
end

There are actually 3 separate snippets of functionality that need testing here, however it is not immediately obvious from the example. Lets be slightly more verbose about what is actually happening:

class MyClass
  def lazy_initialized_value
    @lazy_initialized_value = Expensive.request unless @lazy_initialized_value
    @lazy_initialized_value
  end
end

Now it is much easier to see the 3 steps the code should take:

* Store result of expensive request in instance variable
* Leave instance variable alone when it is already set
* Return the value of the instance variable

Now we have this information, our tests become (using Mocha to mock external methods):

class Expensive; end
 
module Tests::MyClass
  # lazy_initialized_value
  # ----------------------
  class LazyInitializedValueTest < Test::Unit::TestCase
    def test_should_respond
      assert_respond_to MyClass.new, :lazy_initialized_value
    end
 
    def test_should_store_result_of_expensive_request_in_instance_variable
      instance = MyClass.new
      Expensive.stubs(:request).with().returns('expensive value')
      instance.lazy_initialized_value
      assert_equal 'expensive value', instance.instance_variable_get('@lazy_initialized_value')
    end
 
    def test_should_return_value_of_instance_varable
      instance = MyClass.new
      instance.instance_variable_set '@lazy_initialized_value', 'the value'
      Expensive.stubs(:request)
      assert_equal 'the value', instance.lazy_initialized_value
    end
 
    def test_should_maintain_existing_instance_variable_value_when_already_set
      instance = MyClass.new
      instance.instance_variable_set '@lazy_initialized_value', 'existing value'
      Expensive.stubs(:request)
      instance.lazy_initialized_value
      assert_equal 'existing value', instance.instance_variable_get('@lazy_initialized_value')
    end
  end
end

Now we have these tests in place, we can go back and refractor the code ’til our heart’s content using all the tricks in the book but by simplifying the problem in the first place, it gives us a solid test suite and the confidence to make changes without breaking functionality.

If you were solving this problem test-first then you wouldn’t (but more likely, shouldn’t) have written the first example until re-factoring stage anyway, however when these shortcuts become engrained in your brain, it’s all too easy to forget what they are _actually_ doing.

So there we go, simplify the initial implementation, get a solid test suite in order, _then_ re-factor.

The Ruby documentation clearly states that the ‘when’ in a ‘case’ statement uses the ‘===’ method for comparing the subject so I couldn’t work out why something like the following wasn’t working:

# Code
class A; end
 
def foo(instance)
  case instance
    when A : 'an A instance'
    else 'not an A instance'
  end
end
 
# Test
a = stub_everything
a.stubs(:===).with(A).returns(true)
assert_equal 'an A instance', foo(a)

So I had a little play around in irb and found the following:

class A; end
a = A.new
a === A #=> false
A === a #=> true

This revealed that I was actually stubbing the wrong side of the operator, I changed this to the following and voila!

# Code
class A; end
 
def foo(instance)
  case instance
    when A : 'an A instance'
    else 'not an A instance'
  end
end
 
# Test
a = stub_everything
A.stubs(:===).with(a).returns(true)
assert_equal 'an A instance', foo(a)

Looks obvious now but certainly wasn’t at the time!

You can grab our plugin to enable “UnitRecord”:http://unit-test-ar.rubyforge.org/ when using “Autotest”:http://rubyforge.org/projects/zentest below:

“http://svn.soniciq.com/public/rails/plugins/iq_autotest”:http://svn.soniciq.com/public/rails/plugins/iq_autotest

By default, running autotest in the Rails directory will run the unit tests. To run the functional tests, do: AUTOTEST='functional' autotest

I hope this saves some people some time!!