Ruby Forum Rails-core > Assigning to the foreign key on a belongs_to association

I've encountered what seems like an odd omission in the behaviour of
belongs_to associations: if you assign to the foreign key attribute,
it doesn't update the associated object (or mark a previously loaded
one as stale) until you explicitly save or reload it.

Let's say we have a Company model, which belongs_to :city :

torchbox = Company.find_by_name('Torchbox')
=> #<Company id: 1, name: "Torchbox", city_id: 1>
torchbox.city
=> #<City id: 1, name: "Oxford">
torchbox.city_id = 2
=> 2
torchbox.city
=> #<City id: 1, name: "Oxford">

(Would have expected City id: 2 to be returned here...)

I've come up with a 6-line patch (see below) and accompanying plugin
to fix this, but I get the feeling that for something so elementary to
have gone unfixed for so long, there must be some perceived good
reason why it works the way it does. If so, what's the reason? If not,
I'll file this on Lighthouse.

- Matt




diff --git a/activerecord/lib/active_record/associations.rb b/
activerecord/lib/active_record/associations.rb
index 0809b27..7135e50 100644
--- a/activerecord/lib/active_record/associations.rb
+++ b/activerecord/lib/active_record/associations.rb
@@ -1094,6 +1094,13 @@ module ActiveRecord
             instance_variable_set(ivar, new_value.nil? ? nil :
association)
           end

+          if association_proxy_class == BelongsToAssociation
+            define_method("#{reflection.primary_key_name}=") do |
target_id|
+              instance_variable_set(ivar, nil)
+              self["#{reflection.primary_key_name}"] = target_id
+            end
+          end
+
           define_method("set_#{reflection.name}_target") do |target|
             return if target.nil? and association_proxy_class ==
BelongsToAssociation
             association = association_proxy_class.new(self,
reflection)
diff --git a/activerecord/test/cases/associations/
belongs_to_associations_test.rb b/activerecord/test/cases/associations/
belongs_to_associations_test.rb
index b8ec911..c130db2 100644
--- a/activerecord/test/cases/associations/
belongs_to_associations_test.rb
+++ b/activerecord/test/cases/associations/
belongs_to_associations_test.rb
@@ -45,6 +45,13 @@ class BelongsToAssociationsTest <
ActiveRecord::TestCase
     assert_equal apple.id, citibank.firm_id
   end

+  def test_foreign_key_assignment
+    signals37 = accounts(:signals37)
+    assert_equal companies(:first_firm), signals37.firm
+    signals37.firm_id = companies(:another_firm).id
+    assert_equal companies(:another_firm), signals37.firm
+  end
+
   def test_no_unexpected_aliasing
     first_firm = companies(:first_firm)
     another_firm = companies(:another_firm)

On Wed, May 7, 2008 at 8:53 PM, Matt Westcott <matthew@west.co.tt> 
wrote:
> torchbox.city
> have gone unfixed for so long, there must be some perceived good
> reason why it works the way it does. If so, what's the reason? If not,
> I'll file this on Lighthouse.

AFAIK There's no explicit reason that this hasn't been fixed.  There
are a few catches though:

* You'll have to make sure your new method plays nice with the
attribute_methods stuff which does caching (not all primary keys are
ints)
* It'll have to work right when people have already defined their own
mutator too.

So when you attach your patch, make sure you have tests for that stuff.


--
Cheers

Koz