[ ruby-Bugs-9410 ] Bignum.to_s fails for certain values and certain bases (1.8.6 regression)

Bugs item #9410, was opened at 2007-03-21 04:38
You can respond by visiting: 
http://rubyforge.org/tracker/?func=detail&atid=1698&aid=9410&group_id=426

Category: Core
Group: 1.8.6
>Status: Closed
>Resolution: Accepted
Priority: 3
Submitted By: Daniel Azuma (dazuma)
Assigned to: Nobody (None)
Summary: Bignum.to_s fails for certain values and certain bases (1.8.6 regression)

Initial Comment:
irb(main):001:0> RUBY_VERSION
=> "1.8.6"
irb(main):002:0> RUBY_PLATFORM
=> "i686-darwin8.8.1"
irb(main):003:0> 18446744073709551615.to_s(32) # should be "fvvvvvvvvvvvv"
=> "vvvvvvvvvvvv"
irb(main):004:0> 18446744073709551616.to_s(32) # correct
=> "g000000000000"
irb(main):005:0> 18446744073709551615.to_s(36) # should be "3w5e11264sgsf"
=> "w5e11264sgsf"
irb(main):006:0> 18446744073709551616.to_s(36) # correct
=> "3w5e11264sgsg"
irb(main):005:0> 18446744073709551615.to_s(31) # correct
=> "nd075ib45k86f"
irb(main):006:0> 18446744073709551616.to_s(31) # correct
=> "nd075ib45k86g"

I verified this on i686-linux (Fedora Core 5) also.

18446744073709551616 is, of course, 2^64. At first glance, it looks like values less than that threshold, when converted to base 32 or above, are being truncated to 12 characters. Could this have been an optimization gone awry?


----------------------------------------------------------------------

Comment By: Nobuyoshi Nakada  (nobu)
Date: 2007-04-03 13:54

Message:
This bug is fixed in the repository.

----------------------------------------------------------------------

Comment By: Daniel Azuma (dazuma)
Date: 2007-03-21 06:54

Message:
Looking at the bignum.c, I believe this is related:

irb(main):017:0> 18446744073709551615.to_s(8)
=> "1777777777777777777777"
irb(main):018:0> -18446744073709551615.to_s(8)
=> "11777777777777777777777"

I only partially understand the algorithm at this point, but I believe the 
computation of "j" needs to be rounded up in lines 648 and 659 of bignum.c. 
e.g.:

648:   j = j / 3 + 1;

659:   j = j / 5 + 1;

This seems to fix the problem on the test cases I've tried.

Also, to clarify, this appears to be a regression. All the cases I've tried work 
correctly in 1.8.5.

----------------------------------------------------------------------

Comment By: Daniel Azuma (dazuma)
Date: 2007-03-21 04:47

Message:
In case architecture matters, this was reproduced on a MacBook Pro (core duo), 
Mac OS X 10.4.8, with gcc i686-apple-darwin8-gcc-4.0.1 (GCC) 4.0.1 (Apple 
Computer, Inc. build 5341), and on a Dell server (32-bit dual-core I believe, but 
I could be mistaken), Fedora 5, with gcc (GCC) 4.1.1 20060525 (Red Hat 
4.1.1-1).

----------------------------------------------------------------------

You can respond by visiting: 
http://rubyforge.org/tracker/?func=detail&atid=1698&aid=9410&group_id=426