Bug #12437: Is it "legal" to call collect! in class initializer?

• Status: Closed
• Assignee: knu (Akinori MUSHA)
• Target version: -
• ruby -v: ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-linux]
• Backport: 2.1: DONTNEED, 2.2: DONTNEED, 2.3: REQUIRED

[vo.x (Vit Ondruch)]

Is there any reason the following script should not work?

#! /usr/bin/ruby

require 'set'

class Categories < Set

  def initialize(categories=[])
    categories.collect! { |category| category } if categories
    super categories
  end

end

categories = Categories.new()
categories += [1, 2, 3]
p categories

categories2 = Categories.new(categories)
p categories2

It fails with stack level too deep (SystemStackError) error and this regression seems to be introduced by r52591.

For details, please take a look at original issue reported here: https://bugzilla.redhat.com/show_bug.cgi?id=1308057

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Assigned
• Assignee set to knu (Akinori MUSHA)
• Backport changed from 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN to 2.1: DONTNEED, 2.2: DONTNEED, 2.3: REQUIRED

[prijutme4ty (Ilya Vorontsov)]

Second call to categories.collect! works over Categories object. Set#collect! tries to create a new set. Categories.collect! in turn tries to create new Categories.

s=Set.new([1,2,3])
s.object_id # => 7618880
s.collect!{|x| x**2 }
s.object_id # => 7618880 (the same)
s # => #<Set: {1, 4, 9}>

class Categories < Set; end
c = Categories.new([1,2,3])
c.object_id # => 7508680
c.collect!{|x| x**2 }
c.object_id # => 7508680 (the same)
c # => #<Categories: {1, 4, 9}>

Problem is that Set#collect! calls #initialize (probably to maintain some Set properties):

class Categories < Set
  def initialize(*args, &block)
    puts "call to #initialize"
    super
  end
end

c = Categories.new([1,2,3])
# => call to #initalize
c.collect!{|x| x ** 2 }
# => call to #initalize

More precisely, Set#collect! creates new Set and then replaces its content with Set#replace. So each time you call a #collect! etc, you create a new object with your customized constructor which recursively calls #collect and .new. Not a simply-to-fix sort of bug.

[valtri (František Dvořák)]

• File collect-workaround.rb collect-workaround.rb added

Actually it could be fixed by reverting this patch:

https://svn.ruby-lang.org/cgi-bin/viewvc.cgi?view=revision&revision=52591

Or it is possible to worked around that in the "Category" class by using custom collect method with the old code.

So the question is, how it is from ruby language point of view?

The revision 52591 is announced as micro-optimization, so it could be safe to remove it...

[knu (Akinori MUSHA)]

• Status changed from Assigned to Closed

Applied in changeset trunk|r60317.

---

Avoid use of self.class.new(self) in Set#collect!

That prevents infinite recursion when a subclass of Set uses
collect! in its constructor.

This should fix [Bug #12437].