Stubbing a method on any_instance twice causes that method to be undefined in later tests

[bgentry]

I encountered a weird issue with 2.6.0.rc2 where I was stubbing a method on any_instance of a model. In the model specs (which were run after the controller specs) my tests were showing that the method was undefined.

from builds_controller_spec.rb:

before do
  Build.any_instance.stub(:already_built?).and_return(false)
end

it "should not create any builds when both builds already exist" do
  Build.any_instance.stub(:already_built?).and_return(true)
  post :create, :build => { :slot => 'something' }
  Build.count.should == 0
end

from build_spec.rb:

  it { should respond_to :already_built? }

If I run the controller spec first, I see this:

 Failure/Error: it { should respond_to :already_built? }
   expected #<Build id: nil ...> to respond to :already_built?

If I run the tests in the opposite order, everything is fine. I've narrowed the cause down to the specific any_instance line in the controller spec because if I comment out that one line, all of the other tests pass.

My assumption is that because I'm stubbing the same method on any_instance twice, it actually deletes the real method when cleaning up the 2nd time after the spec block.

[kaiwren]

Fixed on c42engineering/rspec-mocks@066d2d1 on the issue54 branch

(I'm not sure if the new 'x referenced this issue from a commit' feature sends out email notifications, hence this comment)

[dchelimsky]

@SidU - thanks so much for taking responsibility for your contribution. Very refreshing. Cheers!

[kaiwren]

Thank you :) It's least I should be doing I'd think.