4 ms·

Ask Matz: why the id of nil is 4?

58 points by potomak 15y ago

swatermasysk 15y ago
Much less relevant today with Ruby 1.9.x, but still very interesting.
- potomak 15y ago
  Yes, you're right but, as you observed, my question is driven by curiosity.
- prodigal_erik 15y ago
  Less relevant how? nil.object_id is still 4 in MRI 1.9.2. The article noted that all integers have odd ids, but not why—the LSB is a http://c2.com/cgi/wiki?TagBit http://c2.com/cgi/wiki?TagBit in MRI. I would hope other implementations are permitted to choose different schemes; it would suck if user code began assuming nobody found a use for more than one tag bit.
  - jarin 15y ago
    This situation was most encountered by Rails developers trying to access the ID of an ActiveRecord object. Since they changed the method to object_id in Ruby 1.9, calling id on a nil object should just give a NoMethodError instead of that slightly cryptic warning.
    - prodigal_erik 15y ago
      Ah, that was a mistake. ActiveRecord should have avoided :id until it was renamed in core Ruby, or returned some sentinel (other than nil) whose :id method fails. Assuming Ruby didn't critically depend on :id succeeding for every object, that is.
JoelMcCracken 15y ago
Interesting. I had always assumed it had something to do with word size being 4.
macournoyer 15y ago
Fixnum values (numbers smaller than native one machine word minus 1 bit) are stored in an object pointer. When you do object.id it returns the address stored in that pointer. For Ruby to know that the value in a pointer is a Fixnum (and not a pointer to an address), it will tag the first bit w/ 1 and shift the integer value by one bit. So storing the value 7 will be done like this: (7 << 1) + 1 # => 14 This is why 7.id == 14. You can see where this is implemented right here: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L225 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#... For true, false and nil they are also stored right in the object pointer using special values defined here: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L354-364 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#.... You'll notice that those 3 values in binary format all end with 0b...10. First bit to 1 means Fixnum, second bit to 1 means special value: true, false or nil.
- technomancy 15y ago
  I found it very odd that he would place false before true, but then I realized that's probably just the C mindset leaking through.
  - macournoyer 15y ago
    Since in Ruby false and nil are the only "falsish" values and everything else is true, their values has been chosen to make a boolean test as efficient as possible. See: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L374 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#...
- xentronium 15y ago
  Sorry for nitpicking, but you meant 15, not 14. Other than that, it seems you are 100% right.
  - macournoyer 15y ago
    Oops! Indeed (7 << 1) + 1 # => 15
klochner 15y ago
Nice S.O. post on the topic: http://stackoverflow.com/questions/3430280/ruby-how-does-object-id-assignment-work http://stackoverflow.com/questions/3430280/ruby-how-does-obj...
silentbicycle 15y ago
You may find Gudeman's "Representing Type Information in Dynamically Typed Languages" (http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.4394 http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.43...) interesting - it's an overview of type tagging strategies used in many dynamically typed languages. It's not specific to Ruby (it's from 1993), but gives more context to the methods Ruby uses.
gabrielh 15y ago
Mark Zuckerbergs facebook user id is 4 (https://graph.facebook.com/4 https://graph.facebook.com/4), so its always fun when you have a nil bug and then you think Zuckerberg is using your site.
flgr 15y ago
Hm, I think it's better style to refer to true and false as true and false (their keywords) instead of TRUE and FALSE (constants, could go away in the future?). :) irb(main):001:0» TRUE » true irb(main):002:0» FALSE » false

58 points by potomak 15y ago

swatermasysk 15y ago

Much less relevant today with Ruby 1.9.x, but still very interesting.

potomak 15y ago
Yes, you're right but, as you observed, my question is driven by curiosity.
prodigal_erik 15y ago
Less relevant how? nil.object_id is still 4 in MRI 1.9.2. The article noted that all integers have odd ids, but not why—the LSB is a http://c2.com/cgi/wiki?TagBit http://c2.com/cgi/wiki?TagBit in MRI. I would hope other implementations are permitted to choose different schemes; it would suck if user code began assuming nobody found a use for more than one tag bit.
- jarin 15y ago
  This situation was most encountered by Rails developers trying to access the ID of an ActiveRecord object. Since they changed the method to object_id in Ruby 1.9, calling id on a nil object should just give a NoMethodError instead of that slightly cryptic warning.
  - prodigal_erik 15y ago
    Ah, that was a mistake. ActiveRecord should have avoided :id until it was renamed in core Ruby, or returned some sentinel (other than nil) whose :id method fails. Assuming Ruby didn't critically depend on :id succeeding for every object, that is.

JoelMcCracken 15y ago

Interesting. I had always assumed it had something to do with word size being 4.

macournoyer 15y ago

Fixnum values (numbers smaller than native one machine word minus 1 bit) are stored in an object pointer. When you do object.id it returns the address stored in that pointer. For Ruby to know that the value in a pointer is a Fixnum (and not a pointer to an address), it will tag the first bit w/ 1 and shift the integer value by one bit. So storing the value 7 will be done like this: (7 << 1) + 1 # => 14 This is why 7.id == 14. You can see where this is implemented right here: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L225 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#... For true, false and nil they are also stored right in the object pointer using special values defined here: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L354-364 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#.... You'll notice that those 3 values in binary format all end with 0b...10. First bit to 1 means Fixnum, second bit to 1 means special value: true, false or nil.

technomancy 15y ago
I found it very odd that he would place false before true, but then I realized that's probably just the C mindset leaking through.
- macournoyer 15y ago
  Since in Ruby false and nil are the only "falsish" values and everything else is true, their values has been chosen to make a boolean test as efficient as possible. See: https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#L374 https://github.com/ruby/ruby/blob/trunk/include/ruby/ruby.h#...
xentronium 15y ago
Sorry for nitpicking, but you meant 15, not 14. Other than that, it seems you are 100% right.
- macournoyer 15y ago
  Oops! Indeed (7 << 1) + 1 # => 15

klochner 15y ago

Nice S.O. post on the topic: http://stackoverflow.com/questions/3430280/ruby-how-does-object-id-assignment-work http://stackoverflow.com/questions/3430280/ruby-how-does-obj...

silentbicycle 15y ago

You may find Gudeman's "Representing Type Information in Dynamically Typed Languages" (http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.4394 http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.43...) interesting - it's an overview of type tagging strategies used in many dynamically typed languages. It's not specific to Ruby (it's from 1993), but gives more context to the methods Ruby uses.

gabrielh 15y ago

Mark Zuckerbergs facebook user id is 4 (https://graph.facebook.com/4 https://graph.facebook.com/4), so its always fun when you have a nil bug and then you think Zuckerberg is using your site.

flgr 15y ago

Hm, I think it's better style to refer to true and false as true and false (their keywords) instead of TRUE and FALSE (constants, could go away in the future?). :) irb(main):001:0» TRUE » true irb(main):002:0» FALSE » false