At a quick glance it is probably because after AttributeMethods is
included in ActiveRecord::Base, write_attribute is aliased &
overwridden (eg the change tracking module). You then change
write_attribute on AttributeMethods but it is too late - the aliasing
that occured in Dirty is pointing at the previous implementation. When
you alias a method ruby does keep track of what the aliased method was
at the time alias_method was called, for example:
class Foo
def to_be_aliased
"implementation 1"
end
alias_method :old_implementation, :to_be_aliased
end
class Foo
def to_be_aliased
"implementation 2"
end
end
Thanks for your response. I think you're on to something and I'll have
to take a look a the Dirty module (amongst others). But, even if the
module was aliasing the original write_attribute, I don't see how this
would interfere with my redefinition.
write_attribute is at the top of the call stack so I'd think it's
going to use the most recent definition. Unless some other module
redefined write_attribute within the scope of active AR::Base (as
apposed to including it via a module) -which is possible.
Using your example, I believe the case is more like the following:
class Foo
def to_be_aliased
p "implementation 1"
end
alias_method :old_implementation, :to_be_aliased
end
class Foo
def to_be_aliased
p "implementation 2"
end
end
At a quick glance it is probably because after AttributeMethods is
included in ActiveRecord::Base, write_attribute is aliased &
overwridden (eg the change tracking module). You then change
write_attribute on AttributeMethods but it is too late - the aliasing
that occured in Dirty is pointing at the previous implementation. When
you alias a method ruby does keep track of what the aliased method was
at the time alias_method was called, for example:
class Foo
def to_be_aliased
"implementation 1"
end
alias_method :old_implementation, :to_be_aliased
end
class Foo
def to_be_aliased
"implementation 2"
end
end
Foo.new.old_implementation #=> "implementation 1"
Hi Fred,
Thanks for your response. I think you're on to something and I'll have
to take a look a the Dirty module (amongst others). But, even if the
module was aliasing the original write_attribute, I don't see how this
would interfere with my redefinition.
write_attribute is at the top of the call stack so I'd think it's
going to use the most recent definition. Unless some other module
redefined write_attribute within the scope of active AR::Base (as
apposed to including it via a module) -which is possible.
Using your example, I believe the case is more like the following:
class Foo
def to_be_aliased
p "implementation 1"
end
alias_method :old_implementation, :to_be_aliased
end
class Foo
def to_be_aliased
p "implementation 2"
end
end
Foo.new.to_be_aliased #implementation 2
It involves alias_method_chain, so it's somewhat like this:
module AR
module M
def x; puts "M#x"; end
def self.included(base)
base.alias_method_chain(:x, :y)
end
end
class Base
def x_with_y; puts "x_with_y"; end
p Base.instance_methods(false).sort # does not include "x"
include M
p Base.instance_methods(false).sort # does include "x"
end
end
AR::Base.new.x
After alias_method_chain, AR::Base has an "x" method, so overriding
the one in AR::M won't affect what happens when you call x.
I stripped it down to the bare bones (and then some, perhaps but I
think this represents what happens. See the file dirty.rb, which is
where the alias_method_chain call is.