Feature #13257


Symbol#singleton_class should be undef

Added by tagomoris (Satoshi TAGOMORI) over 5 years ago. Updated about 2 years ago.

Target version:


Objects of some classes doesn't have singleton classes (e.g., Symbol, Integer, Float...).
Symbol#singleton_class raises TypeError.

But Symbol#respond_to?(:singleton_class) returns true, and we cannot know when #singleton_class returns a valid class or raise errors in any way (except for calling it).

I think that such #singleton_class methods should be undef.

Updated by matz (Yukihiro Matsumoto) over 5 years ago

  • Status changed from Open to Feedback

Fundamentally, #respond_to? is a method to check method existence. The class Symbol does have #singleton_class so that it's natural for it to return true, even though #singleton_class always fails.

It seems you want to check the availability of functionality. That means you want to change the meaning of the method.
I do not reject the proposal immediately, but at least it's controversial. Persuade us if you really want to change it.


Updated by tagomoris (Satoshi TAGOMORI) over 5 years ago

I understood that #respond_to? doesn't matter in fact.
What I really want is to undef Symbol#singleton_class (and Integer, Float too).

Updated by tagomoris (Satoshi TAGOMORI) over 5 years ago

  • Subject changed from Symbol#respond_to?(:singleton_class) should be false (but true) to Symbol#singleton_class should be undef

Updated by tagomoris (Satoshi TAGOMORI) over 5 years ago

  • Description updated (diff)

Updated by knu (Akinori MUSHA) over 5 years ago

What's the motivation behind this? You cannot tell if an object has no singleton class just because it doesn't define or respond to #singleton_class anyway. BasicObject is an exception, and I could imagine certain kinds of objects (for delegate/proxy/mocking) might have their reasons they shouldn't undefine #singleton_method. If you have an object, then (class << object; true; end rescue false) would be the best way to test if it has a singleton class.

Updated by tagomoris (Satoshi TAGOMORI) over 5 years ago

My motivation is to know whether there is an class/module in ancestors or not.

I needed it at this patch:
The feature implemented in this patch serialize/deserialize objects in a configured format if the object is an kind of registered classes or modules.
It should consider about the order of inheritance, so it can't use #is_a? method, and it should consider about modules added by #extend, so it should check ancestors of singleton classes.

Updated by tagomoris (Satoshi TAGOMORI) over 5 years ago

After chat with narse, I found that what I really want is the list of classes and modules on method search path of objects (nearly equal to obj.singleton_class.ancestors).
If I can get it without creating/allocating singleton classes, it's nice to me.

I missed to think about BasicObject instances, but it's out of our scope in this use case.

Updated by knu (Akinori MUSHA) over 5 years ago

Is that a use case where you need a method for checking without allocating a singleton class?
As I read it diagonally, it looks like you are always accessing and using the singleton class when an object has one.

Updated by akr (Akira Tanaka) over 5 years ago

I guess a some method such as Kernel#ancestor_modules, which returns all class/modules to search methods, supports your needs directly than undef'ing singleton_class.

Actions #10

Updated by jeremyevans0 (Jeremy Evans) about 2 years ago

  • Backport deleted (2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN)
  • ruby -v deleted (ruby 2.4.0p0 (2016-12-24 revision 57164) [x86_64-darwin15])
  • Tracker changed from Bug to Feature

Also available in: Atom PDF