Issue #13397 has been updated by headius (Charles Nutter).


You would be well-advised to avoid object_id. It does not do what you think it does. By returning a pointer reference into the garbage-collected heap, it's possible for the same object_id to refer to different objects over time.

I have proposed deprecating and eventually removing both object_id and _id2ref, since implementing them safely would largely make them useless: https://bugs.ruby-lang.org/issues/15408

----------------------------------------
Bug #13397: #object_id should not be signed
https://bugs.ruby-lang.org/issues/13397#change-75624

* Author: vo.x (Vit Ondruch)
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: ruby 2.4.0p0 (2016-12-24 revision 57164) [i386-linux]
* Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN
----------------------------------------
It is surprising that #object_id returns signed value. Let me explain show two examples. Working with 32b Ruby (ruby 2.4.0p0 (2016-12-24 revision 57164) [i386-linux]) to make this issue more apparent.

~~~
$ ruby << \EOR
GC.disable
3_000_000.times { p Object.new.inspect }
EOR
"#<Object:0x57d49a5c>"
"#<Object:0x57d499a8>"
"#<Object:0x57d49930>"
"#<Object:0x57d498b8>"

... snip ...

"#<Object:0x828bf164>"
"#<Object:0x828bf0ec>"
"#<Object:0x828bf074>"
"#<Object:0x828beffc>"
"#<Object:0x828bef84>"
^C-:2:in `p': Interrupt
	from -:2:in `block in <main>'
	from -:2:in `times'
	from -:2:in `<main>'
"#<Object:0x8290b1f4>"
~~~

In this example, the "object_id", which is part of the inspect object is unsigned, since it is printed using C sprintf with %p format. There are other libraries, which tries to mimic the output [ [1] ]. The implementation is approximately following:

~~~
$ ruby << \EOR
GC.disable
class A
  DEFAULT_OBJ_ID_STR_WIDTH = 0.size == 4 ? 7 : 14

  def inspect
    id_str = (object_id << 1).to_s(16).rjust(DEFAULT_OBJ_ID_STR_WIDTH, '0') 
    "#<#{self.class.name}:0x#{id_str}>"
  end
end
3_000_000.times { p A.new.inspect }
EOR

"#<A:0x58585428>"
"#<A:0x585852d4>"
"#<A:0x585851bc>"
"#<A:0x5858507c>"
"#<A:0x58584ec4>"
"#<A:0x58584d5c>"
"#<A:0x58584c1c>"
"#<A:0x58584adc>"

... snip ...

"#<A:0x7fff4888>"
"#<A:0x7fff47c0>"
"#<A:0x7fff46f8>"
"#<A:0x7fff4630>"
"#<A:0x7fff4568>"
"#<A:0x7fff44a0>"
"#<A:0x7fff43d8>"
"#<A:0x7fff4310>"
"#<A:0x7fff4248>"
"#<A:0x7fff4180>"
"#<A:0x7fff40b8>"
"#<A:0x-7fffc034>"
"#<A:0x-7fffc110>"
"#<A:0x-7fffc1ec>"
"#<A:0x-7fffc2c8>"
"#<A:0x-7fffc3a4>"
"#<A:0x-7fffc480>"
"#<A:0x-7fffc55c>"
"#<A:0x-7fffc638>"
^C-:10:in `p': Interrupt
	from -:10:in `block in <main>'
	from -:10:in `times'
	from -:10:in `<main>'
~~~ 

And the output is quite surprising to me. Why the object_id should be signed value? It doesn't make any sense to me. Is this implementation wrong or is Ruby wrong?

[1]: https://github.com/ruby-concurrency/concurrent-ruby/issues/547



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>