[PATCH] Polymorphic associations + single table inheritance + *able_type doesn't record the correct object

Patch to correct polymorphic assoc and STI

I've been running it in a 'real world' application here and on acts_as_taggable. Both ork well. I've also made some tests cases. THe problem is: it doesn't go forward with default active record's test cases. I wasn't able to make the proper corrections to the tests cases. But i'm sure this patch is OK (at least logically)

I can append my tests cases in any case.

I was also having troubles with Polymorphic handling and this patch solved it. Now Rails discover the right class and work as desired.

I'm able to reproduce the problens and apply it on 1.1.6. I hadn't check if 1.2 has the same problem.

Replying to otavio:

I'm able to reproduce the problens and apply it on 1.1.6. I hadn't check if 1.2 has the same problem.

yes 1.2.3 has one

The patch seems to work in 1.2.3 as well, even though the line numbers are off quite a bit.

fix for polymorphic STI associations, for rails 1.2.3 (tested!)

the old patch seems to not patch all the files in rails 1.2.3. attached is a diff that works with rails 1.2.3. i tested it, works like a charm. this should really be put into rails-core since lots of people are using STI and might use polymorphic associations as well, wondering why things start to go wrong.

It still doesn't work when using has_many with :as => 'something' and :dependent => :delete_all (or :destroy) in the base class.

The problem is in rails/activerecord/lib/active_record/associations.rb, function configure_dependency_for_has_many(), line 1049:

dependent_conditions += " AND #{reflection.options[:as]}_type = '#{base_class.name}'"

Here the type field is forced to the base class of the object, which is wrong.

As a workaround, having this in the base class works:

has_many :permissions, :as => 'authobject' def destroy

self.permissions.each {|p| p.destroy} super

end

Is it possible to get Rails to not hard code the condition string immediately, but create it for the correct object, just as when following the relationship? Since "self.permissions" work (with the above patch), it's on the right track. I tried to fix it myself, but couldn't find the right class name.

The tested patch fails for has_many :through associations. Here's a cumulative patch containing my fix to has_many :through as well.

Cumulative patch that fixes polymorphic associations for has_many :through as well.

chriseppstein, your patch won't work with trunk, because there is no method 'quote':

from changelog * Rename AR::Base#quote so people can use that name in their models. #3628 [Koz]

Cumulative trunk patch also fixes some calls to 'base_class.name' where 'name' expected in associations.rb

+1

Replying to antonio:

Let's suppose I have some code like this: {{{ class Picture < ActiveRecord::Base belongs_to :attachable, :polymorphic => true end class FamilyPicture < Picture has_one :attachment, :as => :attachable end class SinglePicture < Picture has_one :attachment, :as => :attachable end }}} Adding an Attachment to a FamilyPicture or SinglePicture (both derived from Picture) results in the addressable_type column referring to "Picture" instead of "FamilyPicture".

This is an implementation detail, not a bug.

When the addressable is instantiated, you'll get a FamilyPicture or SinglePicture based on the pictures.type column (NOT the pictures.addressable_type column).

Also, the patch lacks tests demonstrating what is meant to be fixed.

This is an implementation detail, not a bug. When the addressable is instantiated, you'll get a FamilyPicture or SinglePicture based on the pictures.type column (NOT the pictures.addressable_type column).

How does it help to simply declare the bug invalid?

I'll look into my code to see if I'm missing something, but I wasn't able to get polymorphic associations to work with STI at all until applying the patches.

I also don't agree that this is a implementation detail. We had working code on 1.2, then upgraded to edge, and this patch fixed the tests. It might be lacking some polish, but it defiantly works.

I also still consider this a bug. We have a very similar situation with polymorphic associations where one of the alternatives uses inheritance, and that doesn't work at all without these patches. Not because the wrong class name is stored in one of the tables, but because the wrong type is instanciated when following the relationship. The patch and test might not be perfect, but it's still an unresolved bug.

In response to the multiple voices here, I'm reopening the ticket.

#8841

Polymorphic_sti_associations_edge_v2.patch adjusted for Rails 2.1. I'm not Rails expert, so this patch may break things. But in my simple tests it worked.