be_* predicate gives wrong results when requiring delegate before rspec

[eregon]

This bug is really weird, I searched what caused it during a few hours.

It is probably similar to: http://rubyforge.org/tracker/?func=detail&atid=3149&aid=11526&group_id=797
I tried to find more info about it but did not succeed.
Could you give me access to http://rspec.lighthouseapp.com/projects/5645-rspec/tickets/48-11526-rspec-fails-to-dynamically-bind-delegateclass-methods#ticket-48-2 ?

Here is a small reproducing example:

# compare `RUBYOPT=-rdelegate rspec spec.rb` and `rspec spec.rb`
require 'delegate'

class Immediate
  def cached?
    false
  end
end

class Cached < SimpleDelegator
  def initialize(d)
    super d
  end

  def cached?
    true
  end
end

describe 'A' do
  subject{ Cached.new(Immediate.new) }

  it 'b' do
    p subject.cached? # => true
    subject.should be_cached
  end
end

The example fails only when the 'delegate' library is required before 'rspec', on 1.9 (jruby and mri).
You can also observe this problem by running the specs of rspec-expectations:

RUBYOPT=-rdelegate rake spec

The "arbitrary predicate with DelegateClass"/"accesses methods defined in the delegating class (LH[#48])" (spec/rspec/matchers/be_spec.rb#412) example is indeed testing similar behavior.

I guess the problem mainly lies in lib/delegate.rb which changed a lot from 1.8 on 1.9, notably by using BasicObject. There is some dark magic there, like Kernel.dup.

Until the problem is found, I think RSpec should emit a warning (or abort) when 'delegate' is required before 'rspec' to help users hitting this bug.

What do you think?