Bug #11385: `==` with bidirectional/cyclic dependency

• Status: Rejected
• Assignee: core
• Target version: -
• ruby -v: 2.2.1
• Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN, 2.2: UNKNOWN

[allenwq (Allen Wang)]

class Something
  attr_accessor :friend
  def initialize(friend)
    self.friend = friend
  end

  def ==(other)
    friend == other.friend
  end
end


a = Something.new([])
b = Something.new([a])
a.friend = [b]

a == b

The above code returns true on OS X and Linux, by right it should give me a exception of stack level too deep.

( And on windows I can see the expected exception )

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)
• Status changed from Open to Rejected

true is the expected result, and the same with x64-mswin64_120.

[lowjoel (Joel Low)]

Hi Nakada-san:

D:\> ruby -v
ruby 2.2.3p147 (2015-07-04 revision 51143) [x64-mswin64_120]

D:\> ruby test.rb
test.rb:8:in `==': stack level too deep (SystemStackError)
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
         ... 2574 levels...
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:8:in `=='
        from test.rb:17:in `<main>'

It seems to be a problem specific to Windows. The test suite passes (IIRC)

[Eregon (Benoit Daloze)]

Should the result not be false instead?
These two instances each have a different @friend and it cannot be determined if they are equal so "false" seems a much safer answer.

Also, this seems inconsistent to how Comparable#== treat such cases:

class Something
  attr_accessor :friend
  def initialize(friend)
    self.friend = friend
  end
  include Comparable
  def <=>(other)
    friend <=> other
  end
end
a = Something.new([])
b = Something.new([a])
a.friend = [b]
p a == b # false (and not true like above!)

[Hanmac (Hans Mackowiak)]

my version is "ruby 2.3.0dev (2015-07-21 trunk 51319) [x86_64-linux]"

and it does return true for me too

[Eregon (Benoit Daloze)]

#1448 and the specs in
https://github.com/ruby/rubyspec/blob/master/core/array/shared/eql.rb
convinced me this is desirable behavior.

[Eregon (Benoit Daloze)]

• Related to Bug #1448: [patch] Proper handling of recursive arrays added