Changing method visibility changes method ownership, but doesn't on MRI

[myronmarston]

I'm noticing that when you change method visibility on JRuby, it has a side effect that's not present on MRI: it changes what Method#owner returns.

First, here's the behavior on MRI 2.2:

➜  ~  ruby -v
ruby 2.2.0p0 (2014-12-25 revision 49005) [x86_64-darwin12.0]
➜  ~  ruby -e "class A; private_class_method :new; end; puts A.method(:new).owner"
Class
➜  ~  ruby -e "class A; private :inspect; end; puts A.instance_method(:inspect).owner"
Kernel

It looks like this has always been MRI's behavior; on 1.8.7 I see the same thing:

➜  ~  ruby -v
ruby 1.8.7 (2013-12-22 patchlevel 375) [i686-darwin12.5.0]
➜  ~  ruby -e "class A; private_class_method :new; end; puts A.method(:new).owner"
Class
➜  ~  ruby -e "class A; private :inspect; end; puts A.instance_method(:inspect).owner"
Kernel

Compare that to JRuby's behavior:

➜  ~  ruby -v
jruby 1.7.8 (1.9.3p392) 2013-11-14 0ce429e on Java HotSpot(TM) 64-Bit Server VM 1.6.0_65-b14-462-11M4609 [darwin-x86_64]
➜  ~  ruby -e "class A; private_class_method :new; end; puts A.method(:new).owner"
#<Class:A>
➜  ~  ruby -e "class A; private :inspect; end; puts A.instance_method(:inspect).owner"
A

[pirj]

@headius The following rspec-mocks example passes just fine on jruby 9.2.13.0 (2.5.7) 2020-08-03 9a89c94bcc:

          it 'uses the method signature from `#initialize` for arg verification' do
            # Commented it out and it still passes ❤️ 
            # pending "Failing on JRuby due to https://github.com/jruby/jruby/issues/2565" if RSpec::Support::Ruby.jruby?

            subclass = Class.new(klass) do
              private_class_method :new
            end

            prevents(/arguments/) { allow(subclass).to receive(:new).with(1) }
            allow(subclass).to receive(:new).with(1, 2)
          end

Also, as reported originally:

$ ruby --version
jruby 9.2.13.0 (2.5.7) 2020-08-03 9a89c94bcc OpenJDK 64-Bit Server VM 14.0.1+14 on 14.0.1+14 +jit [darwin-x86_64]
$ ruby -e "class A; private_class_method :new; end; puts A.method(:new).owner"
Class
$ ruby -e "class A; private :inspect; end; puts A.instance_method(:inspect).owner"
Kernel

This seems to have been fixed, so I guess this issue can be closed.

Thanks!

[headius]

Thanks for the update! This was probably fixed when we reworked how method tables function to better support unusual super cases.

[eregon]

FWIW, I think the behavior is actually a CRuby bug: https://bugs.ruby-lang.org/issues/18729
The owner should always be the module/class on which the method is defined AFAIK.
I'm not sure how come RSpec depends on what seems to be a CRuby bug, @bjfish will take a look.

[headius]

I think you meant to link to https://bugs.ruby-lang.org/issues/18729.

[eregon]

Oops, indeed, I edited my comment.