Why was the object_id for true and nil changed in ruby2.0?

Question

I came across this ruby object_id allocation question sometime back and then read this awesome article which talks about VALUE and explains why object_id of true, nil and false the way they are. I have been toying with ruby2.0 object_id when I found the apparent change that has been made regarding object_id of true and nil.

forbidden:~$ ruby -v
ruby 2.0.0p0 (2013-02-24 revision 39474) [x86_64-linux]
forbidden:~$
forbidden:~$ irb
irb(main):001:0> true.object_id
=> 20
irb(main):002:0> false.object_id
=> 0
irb(main):003:0> nil.object_id
=> 8
irb(main):004:0> exit
forbidden:~$
forbidden:~$ rvm use 1.9.3
Using /home/forbidden/.rvm/gems/ruby-1.9.3-p392
forbidden:~$ ruby -v
ruby 1.9.3p392 (2013-02-22 revision 39386) [x86_64-linux]
forbidden:~$
forbidden:~$ irb
irb(main):001:0> true.object_id
=> 2
irb(main):002:0> false.object_id
=> 0
irb(main):003:0> nil.object_id
=> 4

tl;dr: The values for true and nil were respectively 2, 4 in 1.9.3 and 1.8.7, but have been changed to 20, 8 in ruby2.0.0 - even though the id of false remains the same i.e. 0 and the ids for Fixnum maintains the same old 2n+1 pattern.

Also, the way Fixnum and Bignum are implemented is still the same in 2.0.0 as the example given in the above mentioned article also runs just the same way it used to:

irb(main):001:0> 
irb(main):002:0* ((2**62)).class
=> Bignum
irb(main):003:0> ((2**62)-1).class
=> Fixnum
irb(main):004:0>

What's the reason behind this object_id change?

Why was this change made? How is this going to help developers?

Answer 1

A look at the Ruby source where these values are defined suggests that this has something to do with “flonums” (also see the commit where this was introduced). A search for ”flonum” came up with a message on the Ruby mailing list discussing it.

This is a technique for speeding up floating point calculations on 64 bit machines by using immediate values for some floating point vales, similar to using Fixnums for integers. The pattern for Flonums is ...xxxx xx10 (i.e. the last two bits are 10, where for fixnums the last bit is 1). The object_ids of other immediate values have been changed to accomodate this change.

You can see this change by looking at the object_ids of floats in Ruby 1.9.3 and 2.0.0.

In 1.9.3 different floats with the same value are different objects:

1.9.3p385 :001 > s = 10.234
 => 10.234 
1.9.3p385 :002 > t = 10.234
 => 10.234 
1.9.3p385 :003 > s.object_id
 => 2160496240 
1.9.3p385 :004 > t.object_id
 => 2160508080 

In 2.0.0 they are the same:

2.0.0p0 :001 > s = 10.234
 => 10.234 
2.0.0p0 :002 > t = 10.234
 => 10.234 
2.0.0p0 :003 > s.object_id
 => 82118635605473626 
2.0.0p0 :004 > t.object_id
 => 82118635605473626