Bug #6784: Test failures related to numeric with x64 mingw

• Status: Closed
• Assignee: h.shirosaki (Hiroshi Shirosaki)
• Target version: 2.0.0
• ruby -v: ruby 2.0.0dev (2012-07-23 trunk 36499) [x64-mingw32]

[h.shirosaki (Hiroshi Shirosaki)]

x64 mingw build has many floating point numeric related failures in test-all.

http://ci.rubyinstaller.org/job/test-ruby-trunk-x64/24/console

2. Failure:
   test_plus(TestBignum) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_bignum.rb:244]:
   <2535301200456458802993406410752> expected but was
   <2.535301200456461e+30>.

3. Failure:
   test_sub(TestBignum) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_bignum.rb:232]:
   <0> expected but was
   <-2.251799813685248e+15>.

4. Failure:
   test_divide(TestFixnum) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_fixnum.rb:156]:
   <2.328306436538698e-10> expected but was
   <(1/4294967296)>.

5. Failure:
   test_pow2(TestFixnum) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_fixnum.rb:184]:
   <1.5258789062500007e-05> expected but was
   <(1/65536)>.

6. Failure:
   test_divmod2(TestFloat) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_float.rb:269]:
   <4294967296> expected but was
   <4294967295>.

7. Failure:
   test_round_with_precision(TestFloat) [C:/Users/Worker/Jenkins/workspace/git-ruby-trunk/test/ruby/test_float.rb:382]:
   <1.1> expected but was
   <1.0999999999999999>.

This seems cause of broken pow() implementation of x64 mingw.
Using powl() instead of pow() will fix this. I attached the patch.

I tested it with gcc version 4.6.1 (tdm64-1) on Win7.

[luislavena (Luis Lavena)]

• Category set to core
• Status changed from Open to Assigned
• Assignee changed from luislavena (Luis Lavena) to nobu (Nobuyoshi Nakada)
• Target version set to 2.0.0

Thank you Hiroshi,

I'm reassigning this ticket to Nobu looking for approval.

CC: Usa, do you agree with Hiroshi conclusion and solution? If so, I will commit.

Thank you.

[luislavena (Luis Lavena)]

• Assignee changed from nobu (Nobuyoshi Nakada) to h.shirosaki (Hiroshi Shirosaki)

=begin
Thank you Usa,

Hiroshi, see comments from Usa at [ruby-core:46694]:

About Hiroshi's patch, I don't think the name "fake_pow" is
a good name.
We should use "rb_w32_" prefix for the published name.
I think the patch is OK except it.

Please commit variant following those suggestions (e.g. rb_w32_pow)
=end

[usa (Usaku NAKAMURA)]

Hello,

In message "[ruby-core:46690] [ruby-trunk - Bug #6784][Assigned] Test failures related to numeric with x64 mingw"
on Jul.24,2012 08:13:27, [e-mail] wrote:

CC: Usa, do you agree with Hiroshi conclusion and solution? If so, I will commit.

I have not set up the 64bit mingw environment yet.
Therefore, I have not checked this.
However, probably, it will be the problem of mingw, since it
has not occurred in mswin.

About Hiroshi's patch, I don't think the name "fake_pow" is
a good name.
We should use "rb_w32_" prefix for the published name.
I think the patch is OK except it.

Regards,

U.Nakamura [e-mail]

• Status changed from Assigned to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r36522.
Hiroshi, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

Fix broken pow() on x64-mingw32

• include/ruby/win32.h (rb_w32_pow): add new function.
  We use powl() instead of broken pow() for x64-mingw32. This workaround
  fixes test failures related to floating point numeric.
  [ruby-core:46686] [Bug #6784]