A pattern for testing class methods in ruby with rspec explicit subjects

1_intro_to_subject.rb

# RSpec's subject method, both implicitly and explicitly set, is useful for
# declaratively setting up the context of the object under test. If you provide a
# class for your describe block, subject will implicitly be set to a new instance
# of this class (with no arguments passed to the constructor). If you want
# something more complex done, such as setting arguments, you can use the
# explicit subject setter, which takes a block.
describe Person do
  context "born 19 years ago" do
    subject { Person.new(:birthdate => 19.years.ago }
    it { should be_eligible_to_vote }
    its(:age) { should == 19 }
    it "should be younger than a 20 year old" do
      twenty_year_old = Person.new(:birthday => 20.years.ago)
      subject.should be_younger_than(twenty_year_old)
    end
  end
end

2_delayed_execution_of_subject.rb

# The subject block is called after any befores. In fact, it is not called until
# it is referenced in the spec - the compact syntaxes of 'it { should }' and
# 'its(:property)' implicitly call subject as needed*. This means any method
# calls in the subject block reference the state of the #system (like your db),
# at the time subject is called.
describe "the oldest person" do
  subject { Person.new(:birthdate => Person.oldest_person.birthdate - 1) }
  it "should be older than the next oldest person" do
    original_oldest_person = Person.new(:birthdate => 100.years.ago)
    subject.should be_older_than(original_oldest_person)
  end
end

# *The latter syntax actual resets subject, the getter, to be the result of the
# property access, allowing for the intriguing abuse of syntax: 'its(:property) {
# subject.should....'

3_subjecting_a_class_method_itself.rb

# subject thus provides a decent syntax for testing method calls directly.
# Sometimes, though, you have methods you want to test almost purely based on
# what is passed into them, with few other concerns of state, for instance, when
# testing class methods. These specs may have a little shared set-up needed, but
# really the true object under test is the method call itself. In these cases it
# makes sense to set that as the subject, and this leads to an effective and
# succinct, if slightly distasteful, pattern for setting the method arguments via
# standard rspec ivar trickery.
describe ".first" do
  before { Person.create_10_people }
  subject { Person.first(@number_of_people) }
  it "should return the specified number of people" do
    @number_of_people = 5
    subject.size.should == 5
  end
  
  it "should return a single person when so specified" do
    @number_of_people = 1
    subject.should be_a(Person)
  end

  it "should default to returning a single person" do
    subject.should be_a(Person)
  end
end

4_using_a_lambda_for_brevity_and_directness.rb

# I was bemoaning the difference between ruby and more functional languages like
# JavaScript, where you do have ways of easily differentiating between a method
# reference and call (parentheses), and how that would make a pattern like this
# easier by letting you set the method as the subject and then call it in the
# spec. Then I realized, of course ruby has a way for creating a callable
# reference like this: a lambda.
describe ".first" do    
  subject { lambda { |number_of_people| Person.first(number_of_people  ) } }
  it { subject.call(5).size.should == 5 }
  it { subject.call(1).should be_a(Person) }
  it { subject.call.should be_a(Person)
end

5_in_conclusion.txt

If I continue to like this pattern enough after more usage I'll probably see if I
can get them to accept a patch to allow passing of arguments directly into
subject itself. That could possibly get a little messy in the code for subject
itself, but behind the scenes gymnastics to get a nice testing syntax is what
rspec is all about, after all.

Thanks, I had forgotten about this. I would rework #3 to use let(), of course.

#4 is exactly what I was looking for. Still no support for it out-of-the-box...

Just use alllow.