From: "Avdi B. Grimm" <avdi@...>
Date: 2001-09-26T09:00:19+09:00
Subject: [ruby-talk:21701] strange behaviors

Hello,

In the process of a frighteningly productive spate of Ruby coding
today, I ran across some seeming quirks in the language.  I'm curious
whether these are bugs, features, known eccentricities, or what, so I
thought I'd post them and see if anyone else is familiar with the
behaviors I encountered.

Quirk #1: Inconsistent name resolution in instance methods

Ruby looks up unqualified names it finds in method bodies in the
method list of 'self' - usually.

For example:

class Foo
  def bar
    @bar
  end

  def bar=(val)
    @bar = val
  end
  
  def implicit_fetch
    bar # Ruby interprets this as Foo#bar, defined above.
  end
end

f = Foo.new
f.bar = "test 1 2 3"
puts f.implicit_fetch 


This yields the expected response:

test 1 2 3

Ruby correctly interprets the bare reference to 'bar' as a call to
Foo.bar. However, this behavior doesn't hold true for 'setter'
methods. If I expand the class above thus:

class Foo
  def bar
    @bar
  end
  
  def bar=(val)
    @bar = val
  end

  def implicit_assign
    bar = "implicitly assigned" # doesn't call bar=()
  end

  def explicit_assign
    self.bar = "explicitly assigned"
  end

  def implicit_fetch
    bar
  end
end


and then I exercise it:

    f = Foo.new
    f.implicit_assign		# fails
    puts f.implicit_fetch
    f.explicit_assign		# works
    puts f.implicit_fetch

I get the following output:

nil
explicitly assigned

As far as I can tell, ruby is interpreting the
line 

     bar = "implicitly assigned"

to mean "assign the string on the right to a local variable named
'bar'". Unlike for the 'getter' method, Ruby doesn't interpret "bar ="
to mean "call the method of self named bar=()". This despite the fact
that calling "f.bar = some_string" /outside/ the class would result in
a call to bar=().  This was startling inconsistency at first; but
as you can see it's easily gotten around by explicitly referencing
"self.bar =" inside a method that wants to call bar=().

However, that's not the end of the story.  While slightly more
cumbersome, the workaround above works fine - for public or protected
attributes. But what about private 'setter' methods?  Private methods
*can't* be explicitly referenced via "self.method_name"; that's what
makes them private. So if I modify the above class to privatize
bar=(), I find that there's no way at all (that I know of) to access
it:

class Foo
  def bar
    @bar
  end
  
  def bar=(val)
    @bar = val
  end
  
  private :bar=

  def implicit_assign
    bar = "implicitly assigned"
  end

  def explicit_assign
    self.bar = "explicitly assigned"
  end

  def implicit_fetch
    bar
  end
end

f = Foo.new
f.implicit_assign
puts f.implicit_fetch
f.explicit_assign
puts f.implicit_fetch


the above code yields the error:

-:19:in `explicit_assign': private method `bar=' called for
#<Foo:0x4025e310> (NameError)
	from -:31
nil

So, as far as I can tell, private setter methods are completely
inaccessible. This seems sub-optimal, but maybe I'm missing something.


Quirk #2: Constants aren't dynamically resolved

Apparently, each class gets it's own private copy of constants defined
within it, unlike class variables which are shared between parent and
child classes. This useful feature is demonstrated with the following
code:

class Parent
  CONSTANT = "foo"

class Child < Parent
  CONSTANT = "bar"
end

puts Parent::CONSTANT # prints "foo"
puts Child::CONSTANT  # prints "bar"

So far, so good. However, when I assume that Ruby will dynamically
figure out which constant to use based on the class of self, I run
into trouble. E.g.:

class Parent
  CONSTANT = "foo"

  def Parent.implicit
    CONSTANT          # always resolves to Parent::CONSTANT
  end

  def Parent.explicit
    self::CONSTANT    # resolves dynamically
  end
end

class Child < Parent
  CONSTANT = "bar"
end


class Parent
  CONSTANT = "foo"

  def Parent.implicit
    CONSTANT
  end

  def Parent.explicit
    self::CONSTANT
  end
end

class Child < Parent
  CONSTANT = "bar"
end

puts Child.implicit # prints "foo"
puts Child.explicit # prints "bar"


The implicit internal reference to CONSTANT always yields the value of
CONSTANT for the class in which the current method was defined.  To
get the value of CONSTANT for the current class, I must explicitly ask
for it with self::CONSTANT.

Again, this is just a minor annoyance; but it seems to go against the
Principle of Least Surprise.


Anyone have any comments on either of these two quirks?

- Avdi Grimm