YAML, ActiveRecord Serialize and Date formats problem

I have the same issue serialising a hash containing a Date object and can confirm that when running this in console: Date.today.to_yaml => "--- !timestamp 02/04/2009\n"

and again in irb Date.today.to_yaml => "--- 2009-02-04\n"

Is this something to do with Rails locals?

I think I know what's going on here and how to fix it. The problem is that rails redefines Date.to_s (somewhere - haven't actually found it, or perhaps didn't understand what I was looking at) so it takes an optional first argument that selects a named formatting option from those defined in:

ActiveSupport::CoreExtensions::Date::Conversion::DATE_FORMATS

If you create an entry in that table with key ":default" it will end up being used by Date.to_s when you don't supply an argument. And the YAML module, understandably (as it is independent of Rails), uses Date.to_s in this fashion. So when you add a :default date format, you actually alter the standard Date class behavior, and you will likely break YAML support for Date objects.

You can fix this with this bit of code (and if I've got this right, something like this should make its way into the Rails codebase):

class Date                    # reopen Date class
  def to_yaml( opts={} )      # modeled after yaml/rubytypes.rb in std library
    YAML::quick_emit( self, opts ) do |out|
      out.scalar( "tag:yaml.org,2002:timestamp", self.to_s(:db), :plain )
    end
  end
end

This uses the already-defined :db format which happens to be exactly what YAML needs (probably a bit safer to define a separate :yaml format that just happens to be identical to the :db format rather than exploiting this coincidence).

This explains Ivan's observation, that YAMLing dates works just fine in a vanilla irb.

I'd like to +1 this patch - I've just run into this problem as well.

+1 to the patch. I found this ticket because the bug described here causes a model that has :serialize column to break if the column contains something with a Date object.

I applied the patch as a monkey patch and it seem to work. Thanks Andy. I would think this could be added to 2-3-stable fairly easily

+1 I ran into this today.

I am not able to reproduce this problem with 2-3-stable.

>> User.create({:changes=>{:my_date=>Date.today}})
  User Create (0.5ms)   INSERT INTO "users" ("name", "created_at", "updated_at", "changes") VALUES(NULL, '2010-06-25 21:21:37', '2010-06-25 21:21:37', '--- 
:my_date: 2010-06-25
')
=> #<User id: 1, name: nil, changes: {:my_date=>Fri, 25 Jun 2010}, created_at: "2010-06-25 21:21:37", updated_at: "2010-06-25 21:21:37">
>> u = User.first
  User Load (0.8ms)   SELECT * FROM "users" LIMIT 1
=> #<User id: 1, name: nil, changes: {:my_date=>Fri, 25 Jun 2010}, created_at: "2010-06-25 21:21:37", updated_at: "2010-06-25 21:21:37">
>> u.changes
=> {:my_date=>Fri, 25 Jun 2010}

ActiveRecord::Schema.define(:version => 20100625210553) do

  create_table "users", :force => true do |t|
    t.string   "name"
    t.text     "changes"
    t.datetime "created_at"
    t.datetime "updated_at"
  end

end
class User < ActiveRecord::Base
  serialize :changes
end

+1 for this patch. This is still an issue in 2.3.8

+1 for this patch, I'm still seeing this issue in 2.3.10 when using ActiveRecord to serialize a has with a date object.

@Andy Lowry... thank you for the Date patch. I was working on a rake task to backup and reload data and this came in very handy.