These extra array operations appear to be unnecessary. Reasoning:

 * It is not necessary to subtract 'id' from the list of copied attributes because record and mem_record are equal, so therefore their id attributes are also equal (so there is no harm in copying it, and this reduces the complexity of the code)
 * It is not necessary to intersect the attribute names, since record and mem_record are equal, so they have the same id and class, so they have the same columns in the database. If record has non-column attributes then it seems reasonable to also copy them onto mem_record (though I am not sure what situation this would ever happen in)

1 files changed, 2 insertions, 8 deletions

diff --git a/activerecord/lib/active_record/associations/collection_association.rb b/activerecord/lib/active_record/associations/collection_association.rb
index 50825a50f0..525ac65722 100644
--- a/activerecord/lib/active_record/associations/collection_association.rb
+++ b/activerecord/lib/active_record/associations/collection_association.rb
@@ -405,14 +405,8 @@ module ActiveRecord
             mem_record = memory.delete(record)
 
             if mem_record
-              # Only try to assign attributes which exist on mem_record
-              shared = mem_record.attribute_names & record.attribute_names
-
-              # Don't try to assign the primary key, or attributes which have changed on mem_record
-              excluded = ["id"] + mem_record.changes.keys
-
-              (shared - excluded).each do |key|
-                mem_record[key] = record[key]
+              (record.attribute_names - mem_record.changes.keys).each do |name|
+                mem_record[name] = record[name]
               end
 
               mem_record