Ruby Marshal.load doesn't keep order of sorted set

Question

I'm saving a SortedSet object in a file using Marshal.dump. The elements in the set are objects as well (that include Comparable and implement the <=> method).

Later on when restoring that object using Marshal.load, the SortedSet that is loaded from the file is not sorted...

Any idea why or how to fix it?

Here is a simplified example that reproduce the problem:

require 'set'
class Foo
  include Comparable

  attr_accessor :num

  def initialize(num)
    @num = num
  end

  def <=>(other)
    num <=> other.num
  end
end

f1 = Foo.new(1)
f2 = Foo.new(2)
f3 = Foo.new(3)

s = SortedSet.new([f2, f1, f3])

File.open('set_test.dump', 'wb') { |f| Marshal.dump(s, f) }

Than, to load the object from the file i use -

File.open('set_test.dump', 'rb') { |f| ls = Marshal.load(f) }

** I'm using Rails 3.2.3 with Ruby 2.1.8

** When loading the dump from the file - do it in a new/seperate rails console (and don't forget to copy-paste the definition of the Foo class :-) )

Answer 1

Reproducing the bug

I could reproduce this behaviour on every Ruby I tried.

# write_sorted_set.rb
require 'set'
class Foo
  include Comparable

  attr_accessor :num

  def initialize(num)
    @num = num
  end

  def <=>(other)
    num <=> other.num
  end
end

f1 = Foo.new(1)
f2 = Foo.new(2)
f3 = Foo.new(3)

s = SortedSet.new([f2, f1, f3])
File.open('set_test.dump', 'wb') { |f| Marshal.dump(s, f) }
p s.to_a

and

# load_sorted_set.rb
require 'set'
class Foo
  include Comparable

  attr_accessor :num

  def initialize(num)
    @num = num
  end

  def <=>(other)
    num <=> other.num
  end
end

ls = Marshal.load(File.binread('set_test.dump'))
p ls.to_a

When launching

ruby write_sorted_set.rb && ruby load_sorted_set.rb

It outputs

[#<Foo:0x000000010cae30 @num=1>, #<Foo:0x000000010cae08 @num=2>, #<Foo:0x000000010cadb8 @num=3>]
[#<Foo:0x0000000089be08 @num=2>, #<Foo:0x0000000089bd18 @num=1>, #<Foo:0x0000000089bc78 @num=3>]

Why?

Comparable isn't used

Using this definition :

class Foo
  attr_accessor :num
  def initialize(num)
    @num = num
  end
end

in load_sorted_set.rb should raise an exception (comparison of Foo with Foo failed (ArgumentError)), but it doesn't. It looks like SortedSet isn't properly initialized by Marshal.load

lib/set.rb

Looking at the sourcecode for SortedSet :

  module_eval {
    # a hack to shut up warning
    alias old_init initialize
  }

and

      module_eval {
        # a hack to shut up warning
        remove_method :old_init
      }

      @@setup = true
    end
  end

  def initialize(*args, &block) # :nodoc:
    SortedSet.setup
    initialize(*args, &block)
  end
end

It looks like SortedSet has been patched to ensure that SortedSet.setup is executed before any SortedSet is initialized.

Marshal.load doesn't seem to know about this.

Solution

SortedSet.setup

You can call

SortedSet.setup

after require 'set' and before Marshal.load

SortedSet.new

You can force a SortedSet initialization with :

ls = SortedSet.new(Marshal.load(File.binread('set_test.dump')))