From: Johan Holmberg <johan556@...>
Date: 2009-02-12T18:49:45+09:00
Subject: [ruby-core:22033] 1.9.1 & Kernel.load(..., true) --> scope problem?

Hi!
I recently tried to use MRI 1.9.1 for one of my existing scripts.
But I soon found that Kernel.load works differently in MRI 1.9.1 (and
broke my script).
Here is a minimal example, showing the problem:

  # file: loader.rb
  load "./loaded.rb", true


  # file: loaded.rb
  class Foo
      def hello
          puts "This is a Foo"
      end
  end

  class Bar
      def hello
          puts "This is a Bar"
          Foo.new.hello          # 'Foo' not found in MRI 1.9.1
      end
  end

  Bar.new.hello

When I run the "loader.rb" script in MRI 1.8.6 and JRuby 1.1.6 (with or without
the "--1.9" switch) I get the following output:

  This is a Bar
  This is a Foo

This is what I expect. But in MRI 1.9.1 I get an error like this:

  loaded.rb:11:in `hello': uninitialized constant
#<Module:0x3dd3a8>::Bar::Foo (NameError)
        from /Users/holmberg/test/ruby/loaded.rb:15:in `<top (required)>'
        from loader.rb:2:in `load'
        from loader.rb:2:in `<main>'

I have used "Kernel.load(..., true)" on several occasions, when I want
to avoid polluting the global namespace. It seems natural that
classes defined in the same file should "find" each other, even when
we are in an anonymous module.

Is this an intentional change, or a bug in MRI 1.9.1 ?

Regards,
/Johan Holmberg