Factory Girl / Capybara deleting records from database mid-test?

Question

Working with RSpec & Capybara, I'm getting an interesting test failure mode which goes away with a few subtle rearrangements of lines in the test case...stuff that shouldn't matter.

I'm developing my own authentication system. It is currently working and I can login/out with the browser and the session works etc etc. However, trying to test this is failing. Something is going on that I don't quite understand, which seems to depend on the order of (seemingly) unrelated calls.

require 'spec_helper'

describe "Sessions" do
  it 'allows user to login' do
    #line one
    user = Factory(:user)
    #For SO, this method hashes the input password and saves the record
    user.password! '2468'

    #line two
    visit '/sessions/index'


    fill_in 'Email', :with => user.email
    fill_in 'Password', :with => '2468'
    click_button 'Sign in'

    page.should have_content('Logged in')
  end
end

As is, that test fails...the login fails. After inserting 'debugger' calls into both the spec and the controller I can see why: the user is not getting inserted into the database as far as the controller is concerned:

Edit adding in ApplicationController

class ApplicationController < ActionController::Base
  helper :all
  protect_from_forgery

  helper_method :user_signed_in?, :guest_user?, :current_user

  def user_signed_in?
    !(session[:user_id].nil? || current_user.new_record?)
  end

  def guest_user?
    current_user.new_record?
  end

  def current_user
    @current_user ||= session[:user_id].nil? ? User.new : User.find(session[:user_id])
  rescue ActiveRecord::RecordNotFound
    @current_user = User.new
    flash[:notice] = 'You\'ve been logged out.'
  end
end


class SessionsController < ApplicationController
  def login
    user = User.where(:email=>params[:user][:email]).first

    debugger ###

    if !user.nil? && user.valid_password?(params[:user][:password])
      #engage session
    else
      #run away
    end
  end

  def logout
    reset_session
    redirect_to root_path, :notice => 'Logget Out.'
  end
end

in the console, at the above breakpoint:

1.9.2 vox@Alpha:~/Sites/website$ rspec spec/controllers/sessions_controller_spec.rb 
/Users/vox/Sites/website/app/controllers/sessions_controller.rb:7
if !user.nil? && user.valid_password?(params[:user][:password])
(rdb:1) irb
ruby-1.9.2-p180 :001 > User.all.count
 => 0 
ruby-1.9.2-p180 :002 > 

However, if I rearrange a few lines in my test, putting line 'two' above line 'one':

describe "Sessions" do
  it 'allows user to login' do
    #line two
    visit '/sessions/index'

    #line one
    user = Factory(:user)
    #For SO, this method hashes the input password and saves the record
    user.password! '2468'


    fill_in 'Email', :with => user.email
    fill_in 'Password', :with => '2468'
    click_button 'Sign in'

    page.should have_content('Logged in')
  end
end

I get this in the console (same breakpoint as above):

1.9.2 vox@Alpha:~/Sites/website$ rspec spec/controllers/sessions_controller_spec.rb 
/Users/vox/Sites/website/app/controllers/sessions_controller.rb:7
if !user.nil? && user.valid_password?(params[:user][:password])
(rdb:1) irb
ruby-1.9.2-p180 :001 > User.all.count
 => 1 

For the sake of brevity I've omitted the full dump of the contents of the user object but I can assure you that the test completes as expected.

This behavior of swapping lines to get the test to pass doesn't really fit well with my idea of what should be going on with these commands and has proven to be quite a bear to my testing in other areas.

Any hints as to what is going on here?

I've scoured google and SO for ideas which present this problem, and there are no shortage of SO questions about RSpec/Capybara and Sessions. Nothing seemed to fit quite right though.

Thanks for looking.

Update

I've added a breakpoint (just before a visit call) and some debugging to the test and come back with this:

(rdb:1) user
#<User id: 1, login_count: 1, email: "[email protected]", encrypted_password: "11f40764d011926eccd5a102c532a2b469d8e71249f3c6e2f8b...", salt: "1313613794">
(rdb:1) User.all
[#<User id: 1, login_count: 1, email: "[email protected]", encrypted_password: "11f40764d011926eccd5a102c532a2b469d8e71249f3c6e2f8b...", salt: "1313613794">]
(rdb:1) next
/Users/vox/Sites/website/spec/controllers/sessions_controller_spec.rb:19
fill_in 'Email', :with => user.email
(rdb:1) User.all
[]

So clearly something along the way that visit does is telling Factory Girl that its done with the user object and so she deletes it?

Edit After inspecting test.log carefully, nothing is issuing any delete. So I'm more or less back to square one.

Answer 1

With the help of the Factory Girl mailing list I've found the issue.

By default RSpec uses transactions to maintain the database in a clean state and each transaction is tied to a thread. Somewhere along the pipeline the visit_page command splits off and the transaction tied to the current thread dies.

The solution is simple: disable transactions.

describe "Sessions" do
  self.use_transactional_fixtures = false

   it 'no longer uses transactions' do
     #whatever you want
  end
end

Update for Rails 5.1

As of Rails 5.1, use_transactional_fixtures is deprecated and should be replaced with use_transactional_tests.

self.use_transactional_tests = false