• tsukikage (11/11) Feb 12 2011 Please see source in attachment.
• bearophile (4/5) Feb 12 2011 http://d.puremagic.com/issues/show_bug.cgi?id=5568
• tsukikage (3/11) Feb 13 2011 For curiosity, would you explain/guess a possible reason for that?
• bearophile (5/6) Feb 13 2011 I leave the answer to Don, I don't know the internals of D BigInts.
• Don (9/25) Feb 18 2011 That's quite a terrible bug in Bigint. It's slipped through testing

tsukikage <tsukikage dmail.com> writes:

Please see source in attachment.
The output is

  M2 M3 M5 M7 M13 M17 M19 M31 M61 M89 M107 M127 M521 M607 M1279 M2203 M2281
M3217 M4253 M4423
*** M9689***
  M9941 M11213 M19937
*** M21701***
  M23209

It missed 2 Mersenne Primes 9689 & 21701.
Is it my program bug or bigint broken?
It seems subtle.

Thank you!

bearophile <bearophileHUGS lycos.com> writes:

tsukikage:

 Is it my program bug or bigint broken?

http://d.puremagic.com/issues/show_bug.cgi?id=5568

Bye,
bearophile

tsukikage <tsukikage dmail.com> writes:

For curiosity, would you explain/guess a possible reason for that?
Thanks!

bearophile wrote:
 tsukikage:
 
 Is it my program bug or bigint broken?

 
 http://d.puremagic.com/issues/show_bug.cgi?id=5568
 
 Bye,
 bearophile

bearophile <bearophileHUGS lycos.com> writes:

tsukikage:

 For curiosity, would you explain/guess a possible reason for that?

I leave the answer to Don, I don't know the internals of D BigInts.

Multi-precision integers are basic bricks used to build many other programs.
How can you be sure they don't contain bugs that break your heavy numeric code?
This is a big problem.

Bye,
bearophile

Don <nospam nospam.com> writes:

tsukikage wrote:
 Please see source in attachment.
 The output is
 
  M2 M3 M5 M7 M13 M17 M19 M31 M61 M89 M107 M127 M521 M607 M1279 M2203 
 M2281 M3217 M4253 M4423
 *** M9689***
  M9941 M11213 M19937
 *** M21701***
  M23209
 
 It missed 2 Mersenne Primes 9689 & 21701.
 Is it my program bug or bigint broken?
 It seems subtle.
 
 Thank you!
 

That's quite a terrible bug in Bigint. It's slipped through testing 
because of a fault in the win32 Phobos makefile (someone disabled 
asserts!). If Phobos is compiled in debug mode, your code causes an 
assert failure inside the bigint code.

It's a straightforward problem with the fast recursive division 
implementation, failing to properly normalize intermediate quotients.
Unfortunately the fix won't make the next release, which happens in a 
few hours.