Object IDs and Fixnums

I’ve been poking around Ruby’s internals, and I found some cool things about the way Ruby represents integers in memory.

Let’s start with the obvious: in Ruby, everything is an object, including integer literals.

irb> 32.class.name
=> "Fixnum"

Every object has an object_id, which you can think of as representing its location in memory, though since Ruby is a VM it doesn’t correspond to a physical memory address.

irb> "hello".object_id
=> 3347120
irb> {:foo => :bar, :halo => :whirled}.object_id
=> 23332010


For most objects, Ruby creates an (effectively random) object_id when the object is instantiated. But some objects, including true, false, and nil, have fixed object_ids that are the same every time the program runs.

irb> false.object_id
=> 0
irb> true.object_id
=> 2
irb> nil.object_id
=> 4

Fixnum object_ids are also, uh, fixed.

irb> 32.object_id
=> 65

But as you can see, the object_id is not the same as the Fixnum itself. But it’s also not hardcoded into the Ruby source like 0 is for false. Instead, it’s calculated – it’s (Fixnum * 2) + 1.

To understand why it’s done this way, let’s take a closer look at how Ruby stores integers.

In languages that are closer to the hardware like C, the size of an integer is the size of the architecture. So on a 32-bit machine, integers are 32 bits. On 64-bit machines, they’re 64 bits.

In VM-based languages, the implementation is free to mess with this convention. In Ruby, on a 32-bit machine, Fixnums are actually 32 bits, but only use 31 bits to represent the number.

This means a Fixnum can hold any number from -((2**30)-1) to (2**30)-1, which is -1073741824 to 1073741824. An integer literal larger or smaller than that is a Bignum.

irb> 1073741823.class.name
=> "Fixnum"
irb> 1073741824.class.name
=> "Fixnum"
irb> 1073741825.class.name
=> "Bignum"
irb> -1073741824.class.name
=> "Fixnum"
irb> -1073741825.class.name
=> "Bignum"

So…what about that other bit?

Here’s where it gets interesting. It’s that last bit that tells us whether we’re looking at the Fixnum itself, or the Fixnum’s object_id. If that last bit is zero, that bit is ignored and the Fixnum is evaluated from the other 31 bits. If that last bit is 1, it uses all 32 bits to get the Fixnum’s object_id.

An example will make this clearer. The Fixnum 4 has an object_id of 9. 4 in binary is 100. When you put that in a Fixnum, it’s got 28 zeroes in front of it for a total of 31 bits.

0000000000000000000000000000100

Now, what happens if you put another 1 at the end to make 32 bits?

00000000000000000000000000001001