Rails Ruby on Rails → 2.3.10

Comments and changes to this ticket

• You flagged this item as spam.
  

  Roland Moriz December 1st, 2009 @ 01:44 PM

  • Tag changed from “activerecord” to “2.3.5, activerecord”

  I can confirm this on my apps after running Rails 2.3.5. Effectively breaks the apps.

• You flagged this item as spam.
  

  Jeff Tucker December 1st, 2009 @ 03:41 PM

  • Tag cleared.

  I can confirm this as well. It seems to only affect polymorphic associations though?

• You flagged this item as spam.
  

  Jeff Tucker December 1st, 2009 @ 04:33 PM

  • Tag set to “2.3.5, activerecord, rspec_rails”

  Found a solution for my case -- this is a (new) problem with rspec-rails, not ActiveRecord. Using mock_model creates an AR mock with standard methods built in. Apparently destroyed? is a standard method in the association process now, so mock_model instances balk when they get pushed into an association. I resolved this by patching rspec-rails' mock.rb:

        def mock_model(model_class, options_and_stubs = {})
          id = options_and_stubs[:id] || next_id
          options_and_stubs = options_and_stubs.reverse_merge({
            :id => id,
            :to_param => id.to_s,
            :new_record? => false,
        :destroyed? => false, # new stub to handle AR associations
            :errors => stub("errors", :count => 0)
          })
          ...
  

• You flagged this item as spam.
  

  PommyTom December 3rd, 2009 @ 08:32 PM

  Jeffs' fix worked for me... Thanks!

• You flagged this item as spam.
  

  Lee December 3rd, 2009 @ 08:42 PM

  This is fixed in rspec-rails with 48e91

  http://github.com/dchelimsky/rspec-rails/commit/48e91f62ffe84bc150f...

• You flagged this item as spam.
  

  Andrew Bortz December 3rd, 2009 @ 10:02 PM

  • Tag changed from “2.3.5, activerecord, rspec_rails” to “2.3.5, activerecord, activeresource, rspec_rails”

  Actually, for us this was because ActiveResource has not been kept in sync with ActiveRecord, and doesn't implement #destroyed?. When a resource is used through an association... bang!

  Just curious: how abandonware is ActiveResource at the moment?

• You flagged this item as spam.
  

  Andrew Bortz December 16th, 2009 @ 09:18 PM

  • Tag changed from “2.3.5, activerecord, activeresource, rspec_rails” to “2.3.5, 2.3.6, activerecord, activeresource, rspec_rails”

  Since this is a regression in 2.3.5, can we expect it to be fixed in .6?

• You flagged this item as spam.
  

  McClain Looney February 17th, 2010 @ 10:42 PM

  i've just run into this using has_many_polymorphs in conjunction with will_paginate and accepts_nested_attrs_for. Bit of a show stopper for us unfortunately. happens when updating an existing record via accept_nested_attributes.

  rails 2.3.5

• 

  José Valim February 21st, 2010 @ 12:45 PM

  Is this a rspec-rails or Rails issue? Could someone please provide a failing test case?

• 

  Ryan Bigg May 25th, 2010 @ 10:54 AM

  • State changed from “new” to “open”

  Please provide more information on this ticket, especially how to duplicate it.

  Is it still broken in 2.3.6?

• You flagged this item as spam.
  

  acechase June 4th, 2010 @ 10:56 PM

  This is still broken as of rails 2.3.8. Unlike some other folks who've reported this problem, we are not using rspec, we use Flexmock. When are seeing this problem after upgrading from rails 2.2.2 to rails 2.3.8. It looks to happen whenever interacting with associations that are on a mocked AR::Base inheriting class. To work around the problem we added the following code into our flexmock_test_helper (ghetto, but works as a short term solution):

  
  class MockBelongsToAssociation < ActiveRecord::Associations::BelongsToAssociation
    # see rails ticket https://rails.lighthouseapp.com/projects/8994/tickets/3527-undefined-method-destroyed-for-activerecordassociationsbelongstoassociation                                                                             
    def destroyed?
      false
    end
  
    def reload
      # No op                                                                                                                                                                                                                             
    end
  end
  
  class ActiveRecord::Associations::BelongsToAssociation
    # see rails ticket https://rails.lighthouseapp.com/projects/8994/tickets/3527-undefined-method-destroyed-for-activerecordassociationsbelongstoassociation                                                                             
    def destroyed?
      false
    end
  end
  
  
  class ActiveRecord::Associations::BelongsToPolymorphicAssociation
    # see rails ticket https://rails.lighthouseapp.com/projects/8994/tickets/3527-undefined-method-destroyed-for-activerecordassociationsbelongstoassociation                                                                             
    def destroyed?
      false
    end
  end
  

  Obviously this hack isn't the greatest since presumably there are times when destroyed? should actually be returning true.

  I don't fully understand what the new destroyed? method is doing and in what context it is being called from... the stack traces weren't incredibly helpful that come from our failed tests. It seems that the easiest way to replicate the problem is to mock and AR class and then attempt to change one of the associations on that class (problem happens with both has_* and belongs_to associations). It seems that the new destroyed? method is unexpected by the existing mocking frameworks...

• You flagged this item as spam.
  

  acechase June 4th, 2010 @ 11:04 PM

  Quick addition:

  I tested this with Flexmock V0.8.3, then after writing the above comment went back and upgraded Flexmock to v0.8.6 and the problem remains.

  Looking at the usage of AR::Base#destroyed?, it looks like this is now valid AR code that mocking frameworks need to be updated to support?

• You flagged this item as spam.
  

  acechase June 4th, 2010 @ 11:14 PM

  Jeez... I apologize for not having all this better coalesced before posting but I now need to add an additional note.

  So, Flexmock has capacity for mocking AR::Base objects, but it doesn't mock the associations hanging off the object. As soon as you interact with an association from a mocked object then you're right back to interacting with the DB. To avoid this (I had forgotten about this part), we (my fellow engineers at glyde.com) add in our own association mocking in our flexmock_test_helper.

  So my previous comment, that the problem is in the mocking frameworks ballpark, probably applies to some frameworks, but for my companies code base I guess it's on us to fix :)

  I guess I'll need to dig into the AR::Base code to see how a mock should properly track whether or not it's associations are destroyed.

• You flagged this item as spam.
  

  Harm Aarts June 8th, 2010 @ 11:37 AM

  • Tag changed from “2.3.5, 2.3.6, activerecord, activeresource, rspec_rails” to “2.3.5, 2.3.6, 2.3.8, activerecord, activeresource, rspec_rails”

  The problem persists with Rails 2.3.8 and rspec-rails 1.3.2 (which includes Lee's patch). I believe the problem is related to ActiveResource.

• You flagged this item as spam.
  

  coffeeaddict_nl June 29th, 2010 @ 02:47 PM

  • Importance changed from “” to “”

  This doesn't only happen on rspec-rails. It happens in real world apps (eg: non test cases) as well. My best bet is that ActiveRecord::Associations::AssociationProxy should be defining a respond_to? along side with the method_missing.

  I will try to figure out how to do this, cause it is starting to bug me more than I can handle...

• 

  Santiago Pastorino June 29th, 2010 @ 06:35 PM

  • Milestone set to “2.3.9”

• You flagged this item as spam.
  

  coffeeaddict_nl June 30th, 2010 @ 07:01 AM

  This looks as if it is very closely related to #1515 and #2378 (still looking for a solution, but this might be of help to someone interested...)

• You flagged this item as spam.
  

  coffeeaddict_nl June 30th, 2010 @ 02:51 PM

  /me bows head in shame.

  It turns out to be user error on my part.

• You flagged this item as spam.
  

  Harm Aarts July 1st, 2010 @ 09:17 AM

  I believe this bug has something to do with this commit:
  http://github.com/rails/rails/commit/d916c62cfc7c59ab6411407a05b946...

  I also begin to doubt if it has anything to do with rspec.

• You flagged this item as spam.
  

  Michael Edgar August 22nd, 2010 @ 01:21 AM

  During a 2.3.4 to 2.3.8 upgrade trial, our tests started failing with this as well.

  The issue arose specifically when testing whether objects were valid that had associated objects we had mocked out. Previously we had (real models changed):

  @author = flexmock(:model, Author)
  @author.should_receive(:valid?).and_return(true)
  assert_valid Comment.new(:author => @author)
  

  Which would pass. Under the newer versions of Rails, it would give the same issue everyone's had: no destroyed? method.

  In later versions of rails (I'm looking at 2.3.8 right now), #association_valid? is now defined at rails/activerecord/lib/active_record/autosave_association.rb:278, which immediately calls destroyed? and marked_for_destruction? on the associated object in question. Thus, if a mocked associated object needs to have its validity checked, it will have to respond to both destroyed? and marked_for_destruction?. We've changed our code to:

  @author = flexmock(:model, Author)
  @author.should_receive(:valid?).and_return(true)
  @author.should_receive(:destroyed?).and_return(true)
  @author.should_receive(:marked_for_destruction?).and_return(true)
  assert_valid Comment.new(:author => @author)
  

  In the necessary places, and this pain point is no longer an issue. This is, in my opinion, not within Rails' scope. Flexmock (and rspec-rails, if this issue is still a problem for that library) should consider including these methods (valid?, destroyed?, marked_for_destruction?) when one uses flexmock(:model, ModelType).

• 

  Jeremy Kemper August 30th, 2010 @ 02:28 AM

  • Milestone changed from “2.3.9” to “2.3.10”

  [bulk edit]

• You flagged this item as spam.
  

  Michael MacDonald September 19th, 2010 @ 03:39 PM

  I'm experiencing the same problem but I'm using Rails 3.0.0 and it's with a Cucumber (0.9.0 in git) feature where I'm using Machinist (1.0.6) to create a new ActiveRecord object which has a belongs_to association with an ActiveResource object. I'm using Dupe (rails3 branch) to stub out this ActiveResource object. I suspect the problem lies with either Machinist or Dupe. After reading this ticket, I added a destroyed? method into my activeresource class in my test environment (as detailed by acechase) and now it all works for me.

• 

  David Trasbo September 23rd, 2010 @ 09:37 AM

  • State changed from “open” to “needs-more-info”

  People seem to report this error to happen under many different types of circumstances.

  We need a minimal way to reproduce this, a failing test, or a patch.

• You flagged this item as spam.
  

  Michael MacDonald October 3rd, 2010 @ 11:23 PM

  • Tag changed from “2.3.5, 2.3.6, 2.3.8, activerecord, activeresource, rspec_rails” to “2.3.5, 2.3.6, 2.3.8, activerecord, activeresource”

  Coming back to this problem again I realise that it has nothing to do with rspec or with my test setup. The problem exists when using a rails app > 2.3.4 to connect to an existing activeresource (regardless what version of rails that activeresource is running).

  Here's the steps I've used to generate the error:

  • Using rails 2.3.8 via rvm:

      mkdir ~/rails_ticket_3527
      cd rails_ticket_3527
      rvm use 1.8.7@rails238 --create
      gem install rails -v=2.3.8
      gem install sqlite3-ruby
      
      rails clients
      cd clients
      rake db:create
      script/generate scaffold Client
      rake db:migrate
      script/server
  

  Now the resource is running on localhost port 3000.

  • Now via your browser or console, create the first client object

  • Using rails 2.3.4 via rvm:

      cd ~/rails_ticket_3527
      rvm use 1.8.7@rails234 --create
      gem install rails -v=2.3.4
      gem install sqlite3-ruby
  
      rails blog234
      cd blog234
      rake db:create
      script/generate scaffold User client_id:integer
      rake db:migrate
      echo 'class Client < ActiveResource::Base; self.site = "http://localhost:3000"; end' > app/models/client.rb
      echo 'class User < ActiveRecord::Base; belongs_to :client; end' > app/models/user.rb
  

  • Now via script/console in the blog234 app:

      user = User.create
      user.client = Client.find(1)
      user.save
  

  • Repeat the same steps for a rails 2.3.5 app.

  Result: It works in rails 2.3.4 but not in rails 2.3.5.

  I hope that helps!

• 

  Ryan Bigg October 9th, 2010 @ 10:12 PM

  • Tag cleared.

  Automatic cleanup of spam.

• 

  Ryan Bigg October 19th, 2010 @ 08:23 AM

  Automatic cleanup of spam.

• 

  Ryan Bigg October 19th, 2010 @ 08:23 AM

  Automatic cleanup of spam.

• 

  José Valim October 20th, 2010 @ 01:34 PM

  • State changed from “needs-more-info” to “stale”

• You flagged this item as spam.
  

  Michael MacDonald October 20th, 2010 @ 02:20 PM

  • Title changed from “Add magic encoding comment to generated files” to “undefined method `destroyed?' for ActiveRecord::Associations::BelongsToAssociation”
• 

  Jeff Kreeftmeijer November 8th, 2010 @ 08:27 AM

  Automatic cleanup of spam.

• You flagged this item as spam.
  

  Tim Morton April 25th, 2011 @ 09:56 PM

  I can confirm a few things about this issue:

  1) It's still present in 2.3.11.
  2) As a few other people have explained above, the problem occurs when you have a relationship between ActiveRecord and ActiveResource models.
  3) Put this code in an initializer to fix the problem:
  

  class ActiveResource::Base
    def destroyed?
      false
    end
  end
  

  I'll see if I can get a real unit test and fix later this week.