ChatGPT解决这个技术问题 Extra ChatGPT

When monkey patching an instance method, can you call the overridden method from the new implementation?

Say I am monkey patching a method in a class, how could I call the overridden method from the overriding method? I.e. Something a bit like super

E.g.

class Foo
  def bar()
    "Hello"
  end
end 

class Foo
  def bar()
    super() + " World"
  end
end

>> Foo.new.bar == "Hello World"
Shouldn't the first Foo class be some other and the second Foo inherit from it?
nope, i'm monkey patching. I was hoping there would be something like super() I could use to call the original method
This is needed when you don't control the creation of Foo and the usage of Foo::bar. So you have to monkey patch the method.

B
BenKoshy

EDIT: It has been 9 years since I originally wrote this answer, and it deserves some cosmetic surgery to keep it current.

You can see the last version before the edit here.

You can’t call the overwritten method by name or keyword. That’s one of the many reasons why monkey patching should be avoided and inheritance be preferred instead, since obviously you can call the overridden method.

Avoiding Monkey Patching

Inheritance

So, if at all possible, you should prefer something like this:

class Foo
  def bar
    'Hello'
  end
end 

class ExtendedFoo < Foo
  def bar
    super + ' World'
  end
end

ExtendedFoo.new.bar # => 'Hello World'

This works, if you control creation of the Foo objects. Just change every place which creates a Foo to instead create an ExtendedFoo. This works even better if you use the Dependency Injection Design Pattern, the Factory Method Design Pattern, the Abstract Factory Design Pattern or something along those lines, because in that case, there is only place you need to change.

Delegation

If you do not control creation of the Foo objects, for example because they are created by a framework that is outside of your control (like for example), then you could use the Wrapper Design Pattern:

require 'delegate'

class Foo
  def bar
    'Hello'
  end
end 

class WrappedFoo < DelegateClass(Foo)
  def initialize(wrapped_foo)
    super
  end

  def bar
    super + ' World'
  end
end

foo = Foo.new # this is not actually in your code, it comes from somewhere else

wrapped_foo = WrappedFoo.new(foo) # this is under your control

wrapped_foo.bar # => 'Hello World'

Basically, at the boundary of the system, where the Foo object comes into your code, you wrap it into another object, and then use that object instead of the original one everywhere else in your code.

This uses the Object#DelegateClass helper method from the delegate library in the stdlib.

“Clean” Monkey Patching

Module#prepend: Mixin Prepending

The two methods above require changing the system to avoid monkey patching. This section shows the preferred and least invasive method of monkey patching, should changing the system not be an option.

Module#prepend was added to support more or less exactly this use case. Module#prepend does the same thing as Module#include, except it mixes in the mixin directly below the class:

class Foo
  def bar
    'Hello'
  end
end 

module FooExtensions
  def bar
    super + ' World'
  end
end

class Foo
  prepend FooExtensions
end

Foo.new.bar # => 'Hello World'

Note: I also wrote a little bit about Module#prepend in this question: Ruby module prepend vs derivation

Mixin Inheritance (broken)

I have seen some people try (and ask about why it doesn’t work here on StackOverflow) something like this, i.e. includeing a mixin instead of prepending it:

class Foo
  def bar
    'Hello'
  end
end 

module FooExtensions
  def bar
    super + ' World'
  end
end

class Foo
  include FooExtensions
end

Unfortunately, that won’t work. It’s a good idea, because it uses inheritance, which means that you can use super. However, Module#include inserts the mixin above the class in the inheritance hierarchy, which means that FooExtensions#bar will never be called (and if it were called, the super would not actually refer to Foo#bar but rather to Object#bar which doesn’t exist), since Foo#bar will always be found first.

Method Wrapping

The big question is: how can we hold on to the bar method, without actually keeping around an actual method? The answer lies, as it does so often, in functional programming. We get a hold of the method as an actual object, and we use a closure (i.e. a block) to make sure that we and only we hold on to that object:

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  old_bar = instance_method(:bar)

  define_method(:bar) do
    old_bar.bind(self).() + ' World'
  end
end

Foo.new.bar # => 'Hello World'

This is very clean: since old_bar is just a local variable, it will go out of scope at the end of the class body, and it is impossible to access it from anywhere, even using reflection! And since Module#define_method takes a block, and blocks close over their surrounding lexical environment (which is why we are using define_method instead of def here), it (and only it) will still have access to old_bar, even after it has gone out of scope.

Short explanation:

old_bar = instance_method(:bar)

Here we are wrapping the bar method into an UnboundMethod method object and assigning it to the local variable old_bar. This means, we now have a way to hold on to bar even after it has been overwritten.

old_bar.bind(self)

This is a bit tricky. Basically, in Ruby (and in pretty much all single-dispatch based OO languages), a method is bound to a specific receiver object, called self in Ruby. In other words: a method always knows what object it was called on, it knows what its self is. But, we grabbed the method directly from a class, how does it know what its self is?

Well, it doesn’t, which is why we need to bind our UnboundMethod to an object first, which will return a Method object that we can then call. (UnboundMethods cannot be called, because they don’t know what to do without knowing their self.)

And what do we bind it to? We simply bind it to ourselves, that way it will behave exactly like the original bar would have!

Lastly, we need to call the Method that is returned from bind. In Ruby 1.9, there is some nifty new syntax for that (.()), but if you are on 1.8, you can simply use the call method; that’s what .() gets translated to anyway.

Here are a couple of other questions, where some of those concepts are explained:

How do I reference a function in Ruby?

Is Ruby’s code block same as C♯’s lambda expression?

“Dirty” Monkey Patching

alias_method chain

The problem we are having with our monkey patching is that when we overwrite the method, the method is gone, so we cannot call it anymore. So, let’s just make a backup copy!

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  alias_method :old_bar, :bar

  def bar
    old_bar + ' World'
  end
end

Foo.new.bar # => 'Hello World'
Foo.new.old_bar # => 'Hello'

The problem with this is that we have now polluted the namespace with a superfluous old_bar method. This method will show up in our documentation, it will show up in code completion in our IDEs, it will show up during reflection. Also, it still can be called, but presumably we monkey patched it, because we didn’t like its behavior in the first place, so we might not want other people to call it.

Despite the fact that this has some undesirable properties, it has unfortunately become popularized through AciveSupport’s Module#alias_method_chain.

An aside: Refinements

In case you only need the different behavior in a few specific places and not throughout the whole system, you can use Refinements to restrict the monkey patch to a specific scope. I am going to demonstrate it here using the Module#prepend example from above:

class Foo
  def bar
    'Hello'
  end
end 

module ExtendedFoo
  module FooExtensions
    def bar
      super + ' World'
    end
  end

  refine Foo do
    prepend FooExtensions
  end
end

Foo.new.bar # => 'Hello'
# We haven’t activated our Refinement yet!

using ExtendedFoo
# Activate our Refinement

Foo.new.bar # => 'Hello World'
# There it is!

You can see a more sophisticated example of using Refinements in this question: How to enable monkey patch for specific method?

Abandoned ideas

Before the Ruby community settled on Module#prepend, there were multiple different ideas floating around that you may occasionally see referenced in older discussions. All of these are subsumed by Module#prepend.

Method Combinators

One idea was the idea of method combinators from CLOS. This is basically a very lightweight version of a subset of Aspect-Oriented Programming.

Using syntax like

class Foo
  def bar:before
    # will always run before bar, when bar is called
  end

  def bar:after
    # will always run after bar, when bar is called
    # may or may not be able to access and/or change bar’s return value
  end
end

you would be able to “hook into” the execution of the bar method.

It is however not quite clear if and how you get access to bar’s return value within bar:after. Maybe we could (ab)use the super keyword?

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  def bar:after
    super + ' World'
  end
end

Replacement

The before combinator is equivalent to prepending a mixin with an overriding method that calls super at the very end of the method. Likewise, the after combinator is equivalent to prepending a mixin with an overriding method that calls super at the very beginning of the method.

You can also do stuff before and after calling super, you can call super multiple times, and both retrieve and manipulate super’s return value, making prepend more powerful than method combinators.

class Foo
  def bar:before
    # will always run before bar, when bar is called
  end
end

# is the same as

module BarBefore
  def bar
    # will always run before bar, when bar is called
    super
  end
end

class Foo
  prepend BarBefore
end

and

class Foo
  def bar:after
    # will always run after bar, when bar is called
    # may or may not be able to access and/or change bar’s return value
  end
end

# is the same as

class BarAfter
  def bar
    original_return_value = super
    # will always run after bar, when bar is called
    # has access to and can change bar’s return value
  end
end

class Foo
  prepend BarAfter
end

old keyword

This idea adds a new keyword similar to super, which allows you to call the overwritten method the same way super lets you call the overridden method:

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  def bar
    old + ' World'
  end
end

Foo.new.bar # => 'Hello World'

The main problem with this is that it is backwards incompatible: if you have method called old, you will no longer be able to call it!

Replacement

super in an overriding method in a prepended mixin is essentially the same as old in this proposal.

redef keyword

Similar to above, but instead of adding a new keyword for calling the overwritten method and leaving def alone, we add a new keyword for redefining methods. This is backwards compatible, since the syntax currently is illegal anyway:

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  redef bar
    old + ' World'
  end
end

Foo.new.bar # => 'Hello World'

Instead of adding two new keywords, we could also redefine the meaning of super inside redef:

class Foo
  def bar
    'Hello'
  end
end 

class Foo
  redef bar
    super + ' World'
  end
end

Foo.new.bar # => 'Hello World'

Replacement

redefining a method is equivalent to overriding the method in a prepended mixin. super in the overriding method behaves like super or old in this proposal.


@Jörg W Mittag, is the method wrapping approach thread safe? What happens when two concurrent threads call bind on the same old_method variable?
@KandadaBoggu: I'm trying to figure out what exactly you mean by that :-) However, I'm pretty sure it is no less thread-safe than any other kind of metaprogramming in Ruby. In particular, every call to UnboundMethod#bind will return a new, different Method, so, I don't see any conflict arising, regardless of whether you call it twice in a row or twice at the same time from different threads.
Was looking for an explanation on patching like this ever since I started on ruby and rails. Great answer! The only thing missing for me was a note on class_eval vs. reopening a class. Here it is: stackoverflow.com/a/10304721/188462
Where do you find old and redef? My 2.0.0 doesn't have them. Ah, it's hard not to miss the Other competing ideas that did not make it into Ruby were:
V
Veger

Take a look at aliasing methods, this is kind of renaming the method to a new name.

For more information and a starting point take a look at this replacing methods article (especially the first part). The Ruby API docs, also provides (a less elaborate) example.


r
rplaurindo

The class that will make override must to be reloaded after class that contains the original method, so require it in the file that will make overrride.