Ruby

In the ancestors, M is before A, so this seems like a bug:

B.ancestors # => [B, M, A, Object, Kernel, BasicObject] 

Also note that with B.include M, which reports the same ancestors, it works as expected:

module M FOO = 'm' end class A FOO = 'a' end class B < A def foo FOO end end b = B.new p b.foo # => "a" B.include M # instead of A.prepend M p b.foo # => "m" 

This is because unlike method tables, constant tables are not moved to origin classes. This behavior has been present since origin classes were added in Ruby 2.0. I submitted a pull request to work around this: https://github.com/ruby/ruby/pull/4538

jeremyevans0 (Jeremy Evans) wrote in #note-3:

This is because unlike method tables, constant tables are not moved to origin classes.

Do we know if this was somehow intentional or more like an oversight/incorrect optimization? I would think the latter.

This behavior has been present since origin classes were added in Ruby 2.0.

Weren't origin classes introduced at the same time than prepend, specifically to implement prepend?

I submitted a pull request to work around this: https://github.com/ruby/ruby/pull/4538

Thanks!

Eregon (Benoit Daloze) wrote in #note-4:

jeremyevans0 (Jeremy Evans) wrote in #note-3:

This is because unlike method tables, constant tables are not moved to origin classes.

Do we know if this was somehow intentional or more like an oversight/incorrect optimization? I would think the latter.

That I do not know. Most of the talk when prepend was added was about method lookup, but the same issues apply to constant lookup.

This behavior has been present since origin classes were added in Ruby 2.0.

Weren't origin classes introduced at the same time than prepend, specifically to implement prepend?

Yes.

I agree it should be a bug. @nobu (Nobuyoshi Nakada), could you review @jeremyevans's patch? I'd like to see if the fix causes any compatibility issue.

Matz.