Class destruction (evil genius metaprogramming)


#1

I want to write a module which, when included in another module,
destroys every method of a class in that module. I realize this isn’t
really a very useful thing to do, but I want to do it to see if it can
be done. I want to actually include the module and have it replace all
existing methods of this class with just one method.

In Python you can change an object’s class in the middle of your
program by reassigning its magical class variable. That would be a
very simple way to implement this idea. Doesn’t seem possible here,
though.

Here’s an attempt which failed:

class Muppet
def show
“it’s the muppet show!”
end
end
=> nil

kermit = Muppet.new
=> #Muppet:0x123129c

kermit.show
=> “it’s the muppet show!”

class Muppet
remove_method(:show)
end
=> Muppet

kermit.show
NoMethodError: undefined method `show’ for #Muppet:0x123129c
from (irb):36
from :0

So far so good, but I want to get rid of every method. Doing what I
just did programmatically, iterating over all available methods,
that’s the thing which still eludes me.

None of these attempts work:

kermit.methods.each{|m| class << Muppet ; (remove_method(m)) ; end}
NameError: undefined local variable or method m' for #<Class:Muppet> from (irb):43 from (irb):43:ineach’
from (irb):43
from :0

kermit.methods.each{|m| class Muppet; remove_method(m.to_sym) ; end}
NameError: undefined local variable or method m' for Muppet:Class from (irb):44 from (irb):44:ineach’
from (irb):44
from :0

kermit.methods.each{|m| Muppet.class_eval(remove_method(m))}
NoMethodError: undefined method remove_method' for #<Object:0x349f4> from (irb):45 from (irb):45:ineach’
from (irb):45
from :0

I have to admit, it’s not necessarily a bad thing if this task proves
impossible, but I feel like it should be possible.


Giles B.

Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org


#2

On 6/14/07, Giles B. removed_email_address@domain.invalid wrote:

kermit.show
So far so good, but I want to get rid of every method. Doing what I
from :0
from (irb):45
from :0

I have to admit, it’s not necessarily a bad thing if this task proves
impossible, but I feel like it should be possible.

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method #{m.inspect}”)}

Note 1. kermit.methods will return all methods inherited or not.
Note 2. remove method is a class method, so it’s evaluated in the
context of the class object, hence Muppet.instance_eval
Note 3: you could alternatively pass symbols for the method names

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method :#{m}”)}


Rick DeNatale

My blog on Ruby
http://talklikeaduck.denhaven2.com/

IPMS/USA Region 12 Coordinator
http://ipmsr12.denhaven2.com/

Visit the Project Mercury Wiki Site
http://www.mercuryspacecraft.com/


#3

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method #{m.inspect}”)}

Note 1. kermit.methods will return all methods inherited or not.
Note 2. remove method is a class method, so it’s evaluated in the
context of the class object, hence Muppet.instance_eval
Note 3: you could alternatively pass symbols for the method names

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method :#{m}”)}

Muahahaha! That was awesome! Soon the world will quake in fear when I
put this information to use!

But why doesn’t this work?

kermit.methods(false).each do |m|
?> Object.instance_eval(“remove_method :#{m}”)

end
=> []

kermit.methods
=> 53

I tried it with Kernel also and no dice.

Tried it with something else, though, and got a great error message:

(eval):1: warning: removing `initialize’ may cause serious problem

Muahahaha!


Giles B.

Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org


#4

Hi –

On Fri, 15 Jun 2007, Rick DeNatale wrote:

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method #{m.inspect}”)}

Note 1. kermit.methods will return all methods inherited or not.
Note 2. remove method is a class method, so it’s evaluated in the
context of the class object, hence Muppet.instance_eval
Note 3: you could alternatively pass symbols for the method names

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method :#{m}”)}

You can also just use the block form of instance_eval:

Muppet.instance_eval { remove_method(m) }

David


#5

Why should this inspire fear?


#6

On 6/15/07, removed_email_address@domain.invalid removed_email_address@domain.invalid wrote:

Note 3: you could alternatively pass symbols for the method names

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method :#{m}”)}

You can also just use the block form of instance_eval:

Muppet.instance_eval { remove_method(m) }

or Muppet.send :remove_method, :m

Cheers
Robert


#7

Hi –

On Fri, 15 Jun 2007, Giles B. wrote:

=> 53
kermit.methods(false) will give you kermit’s singleton methods:

irb(main):019:0> s = “”
=> “”
irb(main):020:0> def s.x; end
=> nil
irb(main):021:0> s.methods(false)
=> [“x”]

Also, you can’t remove these methods from Object because they’re not
defined there. You’d have to do:

class << s
remove_method(:x)
end

David


#8

Hi –

On Fri, 15 Jun 2007, Robert D. wrote:

context of the class object, hence Muppet.instance_eval
Note 3: you could alternatively pass symbols for the method names

kermit.class.instance_methods(false).each{|m|
Muppet.instance_eval(“remove_method :#{m}”)}

You can also just use the block form of instance_eval:

Muppet.instance_eval { remove_method(m) }

or Muppet.send :remove_method, :m

If you have a method called “m” :slight_smile:

David


#9

Code injection attack to own a RoR site possibly? That would be my
guess.

  • Nathan

#10

On 6/15/07, removed_email_address@domain.invalid removed_email_address@domain.invalid wrote:

Muppet.instance_eval(“remove_method #{m.inspect}”)}

Muppet.instance_eval { remove_method(m) }

or Muppet.send :remove_method, :m

If you have a method called “m” :slight_smile:
Ah I see :(, Probably 50% of my errors come from spurious “:” ARRRRG

Muppet.send :remove_method, m ### TESTED :wink:

Sorry.
Robert


#11

Why should this inspire fear?

Code injection attack to own a RoR site possibly? That would be my guess.

Sort of. I’m trying to use it to build a Rails plugin called
acts_as_fox, which overrides every method on a model to return “chunky
bacon!” (It’s not really that terrifying, I just kind of had a Dr.
Frankenstein moment, drunk on power type thing.)

Unfortunately, applying Rick’s code directly to an ActiveRecord model
doesn’t quite accomplish this, because it’s missing the superclass
methods, but applying it to ActiveRecord::Base doesn’t work either. I
did get it to work by doing it twice, both on the actual model and on
ActiveRecord::Base itself, but that’s very unsatisfying, because I
solved the problem by cutting and pasting. (I think I understand why
it worked; ActiveRecord::Base attaches a lot of methods to its
subclasses, instead of having them inherited directly.) It also fails
to really accomplish what I want to do, because it means that making
one model acts_as_fox destroys all the other models. (I also need to
attach a method_missing to return “chunky bacon!” but that part’s
obviously trivial.)

Really the quickest way to accomplish this would be to simply pop the
model out of its inheritance hierarchy - redefine the model not to
have any superclass except Object - but I don’t know if that’s
possible. Trying it in the most obvious way (class Foo < Object; end,
where Foo was already defined Foo < ActiveRecord::Base) results in a
TypeError with the message “superclass mismatch.”

But there must be a clean way to open up the class, grab all its
methods, including those derived from superclasses, and simply
reassign them. Something like

Foo.instance_methods(true).each{|m| Foo.instance_eval("alias
:chunky_bacon " + m)}


Giles B.

Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org


#12

Hi –

On Sat, 16 Jun 2007, Giles B. wrote:

doesn’t quite accomplish this, because it’s missing the superclass

Really the quickest way to accomplish this would be to simply pop the
model out of its inheritance hierarchy - redefine the model not to
have any superclass except Object - but I don’t know if that’s
possible. Trying it in the most obvious way (class Foo < Object; end,
where Foo was already defined Foo < ActiveRecord::Base) results in a
TypeError with the message “superclass mismatch.”

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

But there must be a clean way to open up the class, grab all its
methods, including those derived from superclasses, and simply
reassign them. Something like

Foo.instance_methods(true).each{|m| Foo.instance_eval("alias
:chunky_bacon " + m)}

Here’s a little demo that does pretty much that (sparing the _ methods
like send):

class Object
def mask
puts “I’m masking a method”
end
end

class C
def x
puts “x”
end

def y
puts “y”
end
end

class D < C
def z
puts “z”
end
end

class D
instance_methods.reject {|m| /^_/.match(m) }.each do |m|
alias_method m, :mask
end
end

C.new.x # x
D.new.x # I’m masking a method
D.new.z # I’m masking a method

David


#13

removed_email_address@domain.invalid wrote:

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

You mean, like Object#become? :slight_smile:

Seriously, how would that play with the core classes, which don’t keep
their state in instance variables?

Devin


#14

On 6/16/07, removed_email_address@domain.invalid removed_email_address@domain.invalid wrote:

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

Actually the ancestry array only exists as the return value of the
ancestry method, internally it’s a chain of class objects, and module
wrapper’s. Some things, like including a module will alter that
chain, although there are restrictions which lead to things like the
double module inclusion problem.

Perhaps evil.rb lets you monkey with the chain, it would seem to be
within its mission.


Rick DeNatale

My blog on Ruby
http://talklikeaduck.denhaven2.com/


#15

Hi –

On Sun, 17 Jun 2007, Rick DeNatale wrote:

On 6/16/07, removed_email_address@domain.invalid removed_email_address@domain.invalid wrote:

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

Actually the ancestry array only exists as the return value of the
ancestry method, internally it’s a chain of class objects, and module
wrapper’s.

By making it writeable I meant making it an interface to the chain
itself. Obviously you can do:

String.ancestors << “blah”

but that wasn’t the point :slight_smile:

David


#16

On 6/16/07, removed_email_address@domain.invalid removed_email_address@domain.invalid wrote:

Not exactly. As I understand it, #become involves references changing
from one object to another. I’m thinking of something more like:

module M
end

a = Object.new
class << a
ancestors.unshift(M) # essentially same as a.extend(M)
end

Would you want to be able to switch ordering too? I wonder what kind
of interesting evil could come of that…


#17

Hi –

On Sun, 17 Jun 2007, Devin M. wrote:

removed_email_address@domain.invalid wrote:

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

You mean, like Object#become? :slight_smile:

Not exactly. As I understand it, #become involves references changing
from one object to another. I’m thinking of something more like:

module M
end

a = Object.new
class << a
ancestors.unshift(M) # essentially same as a.extend(M)
end

Seriously, how would that play with the core classes, which don’t keep their
state in instance variables?

I’m not sure how that would matter specifically. What problem are you
envisioning? (Not that the whole thing wouldn’t be a train-wreck
anyway :slight_smile:

David


#18

Here’s a little demo that does pretty much that

gracias!

(sparing the _ methods
like send):

no mercy! no survivors!

class D
instance_methods.reject {|m| /^_/.match(m) }.each do |m|
alias_method m, :mask
end
end

C.new.x # x
D.new.x # I’m masking a method
D.new.z # I’m masking a method

I tried something very similar and it failed, but I can’t recall why.
I’ll check this later in the day.


Giles B.

Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org


#19

I do sometimes wonder what would happen if the ancestry array were
writeable. It could be interesting. I haven’t thought through the
possible pitfalls.

Would you want to be able to switch ordering too? I wonder what kind
of interesting evil could come of that…

You can play with this in Python. If you modify the class magic
var you can change an object’s class in real-time. When I first
discovered this, I posted on a Python list that it seemed like a
sportscar without a seatbelt, and got some very outraged responses
from Python programmers who felt passionately that it was one of the
language’s best features. The most memorable response turned my words
around (“sportscar without a speed limit”).

On the flipside, I also heard a story from a guy who found this kind
of thing heavily abused in some legacy bioinformatics code in Python,
and he was not happy about it. Objects would wander off into
particular methods and come back home completely transformed. It
sounded very confusing.


Giles B.

Blog: http://gilesbowkett.blogspot.com
Portfolio: http://www.gilesgoatboy.org