From: David Masover <ninja@...>
Date: 2009-09-15T10:08:33+09:00
Subject: Re: Overflow behavior

On Sunday 13 September 2009 05:52:03 pm Joel VanderWerf wrote:
> jacob navia wrote:
> > What is the official behavior of ruby in case of integer overflow?
>
> Integers overflow from Fixnums (30 bits) into Bignums (unlimited
> precision integers).

Not entirely accurate, apparently:

irb(main):009:0> (2**62-1).class
=> Fixnum
irb(main):010:0> (2**62).class
=> Bignum
irb(main):012:0> ((-2)**62-1).class
=> Fixnum
irb(main):013:0> ((-2)**62).class
=> Bignum

Fixnum seems to be a signed 63-bit integer (62 bits plus the sign bit) on my 
64-bit system. I'm guessing that it's actually a 31-bit signed integer on a 
32-bit system.

This is nice in that my programs don't really care, but will use a 64-bit 
integer where it's efficient -- yet it strikes me that for smaller numbers, 32 
bits would be more efficient, RAM-wise, if nothing else. Then again, I don't 
know from this (nor should I have to care) how many bits Fixnum actually uses.

Other interesting things: There only seems to be one Float class, and it can't 
go as high as Bignum.

irb(main):033:0> (2**1023).to_f
=> 8.98846567431158e+307
irb(main):034:0> 2**1024
=> 
179769313486231590772930519078902473361797697894230657273430081157732675805500963132708477322407536021120113879871393357658789768814416622492847430639474124377767893424865485276302219601246094119453082952085005768838150682342462881473913110540827237163350510684586298239947245938479716304835356329624224137216
irb(main):035:0> (2**1024).to_f
=> Infinity

And there's your answer for overflow behavior. It seems Bignum integers will 
happily consume all your RAM if you let them, whereas Floats will occasionally 
round up to Infinity.