Bug in Bignum.to_s (1.8.6 only)

[Daniel Azuma]

Hello all,

Just making a quick note of a bug we found in Bignum.to_s in Ruby 1.8.6,
in case it burns anyone else.

Examples:

irb(main):001:0> 18446744073709551615.to_s(36) # should be
"3w5e11264sgsf"
=> "w5e11264sgsf"
irb(main):002:0> -18446744073709551615.to_s(8) # should be
"-1777777777777777777777"
=> "11777777777777777777777"

The bug affects bases 8, 9, and 32-36 inclusive, over certain ranges of
values depending on the base. Values immediately less than 2**64 seem to
be particularly affected. For example, in base 36, all values from
4738381338321616896 through 18446744073709551615 (which is 2**64-1) will
fail.

It manifests only in Ruby 1.8.6. Ruby 1.8.5 works correctly. (It appears
that some optimization work went a little too far.)

For anyone affected by this, I filed the bug on RubyForge, and posted a
possible "works for me, but use at your own risk" fix to bignum.c:

http://rubyforge.org/tracker/?func=detail&aid=9410&group_id=426&atid=1698

Cheers,
Daniel Azuma
Chief Architect, Zoodango.com

 

[bbiker]

Hello all,

Just making a quick note of a bug we found in Bignum.to_s in Ruby 1.8.6,
in case it burns anyone else.

Examples:

irb(main):001:0> 18446744073709551615.to_s(36) # should be
"3w5e11264sgsf"
=> "w5e11264sgsf"
irb(main):002:0> -18446744073709551615.to_s(8) # should be
"-1777777777777777777777"
=> "11777777777777777777777"

The bug affects bases 8, 9, and 32-36 inclusive, over certain ranges of
values depending on the base. Values immediately less than 2**64 seem to
be particularly affected. For example, in base 36, all values from
4738381338321616896 through 18446744073709551615 (which is 2**64-1) will
fail.

It manifests only in Ruby 1.8.6. Ruby 1.8.5 works correctly. (It appears
that some optimization work went a little too far.)

For anyone affected by this, I filed the bug on RubyForge, and posted a
possible "works for me, but use at your own risk" fix to bignum.c:

http://rubyforge.org/tracker/?func=detail&aid=9410&group_id=426&atid=...

Cheers,
Daniel Azuma
Chief Architect, Zoodango.com

I have verified that the bignum bug you described also occurs in ruby
1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

C:\Documents and Settings\Owner>irb
irb(main):001:0> 18446744073709551615.to_s
=> "18446744073709551615"
irb(main):002:0> 18446744073709551615.to_s(32)
=> "vvvvvvvvvvvv"
irb(main):003:0> 18446744073709551616.to_s(32)
=> "g000000000000"
irb(main):004:0> 18446744073709551615.to_s(36)
=> "w5e11264sgsf"
irb(main):005:0> 18446744073709551616.to_s(36)
=> "3w5e11264sgsg"
irb(main):006:0> 18446744073709551615.to_s(31)
=> "nd075ib45k86f"
irb(main):007:0> 18446744073709551616.to_s(31)
=> "nd075ib45k86g"

 

[Daniel Azuma]

I have verified that the bignum bug you described also occurs in ruby
1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

Actually, to clarify, the 1.8.6 releases are the ONLY released version
I'm aware of that exhibits the bug. (1.8.5 and earlier do not appear to
be affected, so far as I know.) But thanks for verifying.

Daniel