[2.1] Refinements

[Charles Oliver Nutter]

Refinements have been pared back to a much more reasonable form that does not have performance implications for unrefined code. They will ship as an official feature in 2.1.

We need to implement refinements in JRuby, at least as an experiment, as soon as possible to know whether they have undiscovered issues. Once they ship in 2.1 it will be much harder to fix them.

I'm marking this for 1.7.6, since I want to do the experiment ASAP.

https://bugs.ruby-lang.org/projects/ruby-trunk/wiki/RefinementsSpec

[Thomas E Enebo]

I realized if I land this in its current 50%+ working state it will not work in JIT since in this form it still has a refinedcallinstr vs using a refinedcallsite. Even had I made that change the JIT would still need some work so I am deferring this until next 9k release.

[Thomas E Enebo]

At the point of pre2 we are passing a substantial portion of refinement tests but we have quite a few significant failing tests (see test/mri). So more work to do before RC

[Mikael Henriksson]

Any progress on this?

[James Adam]

I've hit some strange behaviour trying to use refinements with version 9.0.4.0:

module Refinement
  refine Object do
    def new_method
      'new method!'
    end
  end
end

class Thing
  using Refinement

  p in_class_definition: Object.new.new_method

  def test
    p in_method: Object.new.new_method
  end
end

Thing.new.test

outputs

{:in_class_definition=>"new method!"}
NoMethodError: undefined method `new_method' for #<Object:0xcfa9dbf>
   test at test.rb:15
  <top> at test.rb:19

It seems like there's something amiss with the lexical scoping of the method; in MRI, methods defined with the lexical scope where a refinement is activated can make use of that refinement. Is this expected behaviour in JRuby as of 9.0.4.0?

[Jakub Jirutka]

Another related issue: #2977.

[Charles Oliver Nutter]

I'm going to close this one because refinements are at least in place, albeit rather buggy. #3548 and #2977 track known issues.