Inconsistency between macruby and ruby in a simple class

[kfowler@…]

running this script:

class W
  def initialize(w)
    @w = w
  end
end

class M
  def initialize(m)
    @m = m
  end
end


printf("%s\n" , W.new(W.new(1)).inspect)
printf("%s\n" , M.new(M.new(1)).inspect)
printf("%s\n" , W.new(M.new(1)).inspect)
printf("%s\n" , M.new(W.new(1)).inspect)

on ruby1.9 from mac ports (ruby 1.9.1p243 (2009-07-16 revision 24175) [i386-darwin10]) yields this:

#<W:0x0000010084ecd8 @w=#<W:0x0000010084edf0 @w=1>>
#<M:0x0000010084e370 @m=#<M:0x0000010084e3e0 @m=1>>
#<W:0x0000010084dd88 @w=#<M:0x0000010084de30 @m=1>>
#<M:0x0000010084d618 @m=#<W:0x0000010084d7a0 @w=1>>

on macruby 0.5 (MacRuby version 0.5 (ruby 1.9.0) [universal-darwin10.0, x86_64]) built from source ( http://svn.macosforge.org/repository/ruby/MacRuby/trunk@2771) against llvm ( https://llvm.org/svn/llvm-project/llvm/trunk@82747) yields this:

##-<W:0x20001cc00 @w=1>
##-<M:0x20001cfe0 @m=1>
##-<W:0x20001d260 @w=##-<M:0x20001d2a0 @m=1>>
##-<M:0x2000162a0 @m=##-<W:0x2000162e0 @w=1>>

Ruby 1.8.7 from 10.6:

#<W:0x1001558c8 @w=#<W:0x1001558f0 @w=1>>
#<M:0x100155788 @m=#<M:0x1001557b0 @m=1>>
#<W:0x100155670 @w=#<M:0x100155698 @m=1>>
#<M:0x100155558 @m=#<W:0x100155580 @w=1>>

jruby 1.3.1:

#<W:0x141dddba @w=#<W:0x6d732ed2 @w=1>>
#<M:0x60c9630a @m=#<M:0x4585572a @m=1>>
#<W:0x351daa0e @w=#<M:0x2e879860 @m=1>>
#<M:0x4824de7d @m=#<W:0x6c8b058b @w=1>>

[kfowler@…]

if I change M to extend W: class M < W... then I get:

##-<W:0x2000166c0 @w=1>
##-<M:0x200016aa0 @m=1>
##-<M:0x200016e20 @m=1>
##-<M:0x200010980 @m=##-<W:0x2000109c0 @w=1>>

[kfowler@…]

"return self" appears to be a work around:

  def initialize(m)
    @m = m
    return self
  end

[conradwt@…]

Yes, there's seems to be an issue with the initialize method returning the instance to the caller. However, you can rewrite the classes as follows and the initialize method seems to work as expected:

W = Class.new do
   define_method :initialize do |w|
      @w = w
   end
end

M = Class.new do
   define_method :initialize do |m|
      @m = m
   end
end

printf("%s\n" , W.new(W.new(1)).inspect)
printf("%s\n" , M.new(M.new(1)).inspect)
printf("%s\n" , W.new(M.new(1)).inspect)
printf("%s\n" , M.new(W.new(1)).inspect)

Next, adding 'self' as the last executable statement within the 'initialize' method provides a temporary workaround. Finally, if I'm reading the code correctly, the function 'rb_class_new_instance0' within the 'object.c' file may be a starting point for resolving the issue.

[lsansonetti@…]

Good catch, the problem is definitely in object.c:1896

    if (init_obj != Qnil) {
        p = CLASS_OF(init_obj);
        while (p != 0) {
            if (p == klass) {
                return init_obj;
            }
            p = RCLASS_SUPER(p);
        }
    }

If I remember correctly, this was added because some classes (Array for example) return different objects during #initialize that should be taken into account. The code should probably be altered to only handle these objects.

[lsansonetti@…]

After a quick investigation it looks like this code was not necessary anymore, I deleted it in r2775.

[conradwt@…]

I noticed that the output is printing extra '#-'. For example,

##-<M:0x200010980 @m=##-<W:0x2000109c0 @w=1>>

when it should be represented as

#<M:0x200010980 @m=#<W:0x2000109c0 @w=1>>

[lsansonetti@…]

Indeed, there seems to be a problem in #inspect when the receiver has at least one instance variable:

$ ./miniruby -e "class Foo; def initialize; @foo=42; end; end; p Foo.new.inspect"
"##-<Foo:0x200025d20 @foo=42>"
$ ruby19 -e "class Foo; def initialize; @foo=42; end; end; p Foo.new.inspect"
"#<Foo:0x0000010086a3c0 @foo=42>"

Could you open a new ticket about this?