Could my patch on #3678 please be considered? It's just a small
optimization - currently the @attributes hash is duped (the slow way -
going through attribute_names and injecting into a new hash) on every
record save when there's no normal attributes changed, only to check
if any of the attributes loaded are also in serialized_attributes.
My patch simply changes the method to perform its check using
@attributes rather than calling attributes. This fix speeds up saves
and perhaps more importantly reduces garbage. It has also the
beneficial side effect of not triggering a load of lazy-loaded
attributes if you use a plugin like my columns_on_demand, though
that's not so important for core.