views:

223

answers:

1

It took me a while to understand how private methods work in Ruby, and it really strikes me as being very awkward. Does anyone know if there are good reasons for private methods to be handled the way they are? Is it just historic reasons? Or implementation reasons? Or are there good solid logical reasons (ie. semantic)?

For example:

class Person
  private
  attr_reader :weight
end

class Spy < Person
 private
  attr_accessor :code
 public
  def test
    code          #(1) OK: you can call a private method in self
    Spy.new.code  #(2) ERROR: cannot call a private method on any other object
    self.code     #(3) ERROR!!! cannot call a private method explicitly on 'self'
    code="xyz"    #(4) Ok, it runs, but it actually creates a local variable!!!
    self.code="z" #(5) OK! This is the only case where explicit 'self' is ok
    weight        #(6) OK! You can call a private method defined in a base class
  end
end
  • Ruby's behaviour on lines (1), (2) and (5) seems reasonable.
  • The fact that (6) is ok is a bit strange, especially coming from Java and C++. Any good reason for this?
  • I really do not understand why (3) fails ! An explanation, anyone?
  • The problem on line (4) looks like an ambiguity in the grammar, which has nothing to do with 'private'.

Any ideas?

+14  A: 

You might find it helpful to read ruby's definition of public, private and protected. (Skip to Access Control)

Ruby's private is analogous to Java's protected. There is no Ruby equivalent of Java's private. EDIT: This solution now provides a method of faking it Java's ideal of private in Ruby objects.

Private is defined as methods/variables that can only be called implicitly. This is why statements 2 and 3 fail. In other words, private limits methods/variables to the context of a class or subclass in which they are defined. Inheritance passes private methods to the subclasses and can therefore be accessed with an implicit self. (Explaining why statement 6 works.)

I think you're looking for something closer to protected. Which behaves similarly to Java accessors that are not given a visibility (eg: public, private, protected) By changing the private in Spy to protected all 6 of your statements work. Protected methods can be called by any instance of the defining class or their subclasses. Either explicitly or implicitly called on self are valid statements for protected methods so long as the caller is either the class of the object responding to the call, or inherits from it.

class Person
  private
  attr_reader :weight
end

class Spy < Person
 protected
  attr_accessor :code
 public
  def test
    code          #(1) OK: you can call a private method in self
    Spy.new.code  #(2) OK: Calling protected method on another instance from same class family or a descendant.
    self.code     #(3) OK: Calling protected method on with explicit self is allowed with protected
    code="xyz"    #(4) Ok, it runs, but it actually creates a local variable!!!
    self.code="z" #(5) OK! This is the only case where explicit 'self' is ok
    weight        #(6) OK! You can call a private method defined in a base class
  end
end

s = Spy.new
s.test # succeeds
s.code #(7) Error: Calling protected method outside of the class or its descendants.

As for statement 4. You are correct in assuming this is to avoid ambiguity. It's more a safeguard to the potential harm of ruby's dynamic nature. It ensures that you cannot override accessors by opening up the class again later. A situation that can arise, for example by eval'ing tainted code.

I can only speculate on he design decisions that led to these behaviours. For most of it I feel it comes down to the dynamic nature of the language.

P.S. If you really want to give things the java definition of private. Only available to the class in which it's defined, not even subclasses. You could add an self.inherited method to your classes to remove references to the methods you want to limit access to.

Making the weight attribute inaccessible from subclasses:

class Person
  private
  attr_reader :weight

  def initialize
    @weight = 5
  end

  def self.inherited(subclass)
    subclass.send :undef_method, :weight
  end
end

class Spy < Person
 private
  attr_accessor :code
 public
  def test
     weight       
  end
end

Person.new.send(:weight)  # => 5
Spy.new.send(:weight)  #=> Unhelpful undefined method error

It may make more sense to replace the undef_method call to something like this:

  def self.inherited(subclass)
    subclass.class_eval %{
      def weight 
        raise "Private method called from subclass. Access Denied"
      end
     }
  end

Which provides a much more helpful error and the same functionality.

The send is necessary to get around calling private methods for other classes. Only used to prove that things are actually working.

Which in hindsight, makes private and protected useless. If you're really serious about protecting your methods you will have to override send to block them. The following code does that based on the private_methods of the object:

def send_that_blocks_private_methods(method, *args)
  if private_methods.include?(method.to_s)
    raise "Private method #{method} cannot called be called with send."
  else
    send_that_allows_private_methods(method, *args)
  end
end

alias_method :send_that_allows_private_methods, :send
alias_method :send, :send_that_blocks_private_methods
private :send_that_allows_private_methods

You could specify a class_variable of private_methods you want to block access to instead of denying access to all private methods. You could also make send private, but there are legitimate uses of calling send from outside an object.

EmFi
+1 Thanks for your answer: it is very clear.
MiniQuark