Remove !new_record? requirement from ActiveRecord::Base#==

class Thing < ActiveRecord::Base; end

assert_equal Thing.new, Thing.new

# fails with 
#   <#<Thing id: nil, created_at: nil, updated_at: nil>> expected but was
#   <#<Thing id: nil, created_at: nil, updated_at: nil>>.

This fails because AR::Base includes !comparison_object.new_record? in its definition of ==(comparison_object): http://github.com/rails/rails/blob/c6e20586372743ce200449bf0ac21aed...

This often leads to confusion when trying to test things that involve new records. I brought this up with Jeremy a year or two ago and if memory serves he agreed that removing that requirement would make testing easier but that there were reasons not to do so.

One was the notion of entity, and the idea that an identity-less object is not an entity (although, ironically, if you remove the "id" from "identity", you're left with "entity" :) )

The more pragmatic argument was that if new records were considered equivalent (assuming the other attributes were equal), it would be hard to know how this would impact code in the wild.

I'm bringing this up now in the hopes that we can have a brief discussion of this again before Rails 3 is released. This is the perfect opportunity to take the risk, as all eyes are on upgrade process and documentation.

Thoughts?