Bug #7276: TestFile#test_utime failure

• Status: Closed
• Assignee: nobu (Nobuyoshi Nakada)
• Target version: 2.0.0
• ruby -v: ruby 2.0.0dev (2012-11-04 trunk 37467) [i386-mingw32]

[jonforums (Jon Forums)]

I'm currently travelling and am in UTC-5. I get the following make test-all failure on Win7 32bit with mingw-w64 gcc 4.7.2. It appears that trunk@37467 is not handling daylight savings time correctly.

I'm not going to have time to try to repro on my Arch system, but can anyone else in UTC-5 (or other DST observing region) repro on either a Win7 or Linux system?

[ 3509/11467] TestFile#test_utime = 0.00 s
5) Failure:
test_utime(TestFile) [c:/Jenkins/workspace/ruby-trunk-svn/test/ruby/test_file.rb:287]:
[ruby-core:44776].
<2011-05-27 16:10:39 -0400> expected but was
<2011-05-27 15:10:39 -0400>.
...
Finished tests in 641.449135s, 17.8767 tests/s, 4150.9932 assertions/s.
11467 tests, 2662651 assertions, 1 failures, 0 errors, 83 skips

ruby -v: ruby 2.0.0dev (2012-11-04 trunk 37467) [i386-mingw32]

[jonforums (Jon Forums)]

Appears to be a trunk regression as TestFile#test_utime passes on ruby_1_9_3@37334

C:\Jenkins\workspace\ruby-1_9_3-svn>ruby -v test\ruby\test_file.rb
ruby 1.9.3p305 (2012-10-26 revision 37334) [i386-mingw32]
Run options:

Running tests:

................................

Finished tests in 1.154402s, 27.7200 tests/s, 127.3387 assertions/s.

[mame (Yusuke Endoh)]

• Status changed from Open to Assigned
• Assignee set to nobu (Nobuyoshi Nakada)

[jonforums (Jon Forums)]

Ping.

This is the only test-all fail I have when building trunk on Win7 with mingw-w64 gcc 4.7.2.

This fail does not show up at http://ci.rubyinstaller.org/ because the current build slaves reside in an area that doesn't participate in the daylight savings time circus.

[luislavena (Luis Lavena)]

• Assignee changed from nobu (Nobuyoshi Nakada) to usa (Usaku NAKAMURA)

Nakamura-san,

Do you have any idea about this? Nobu seems not responding so perhaps you can help us figure out the best solution for this.

Thank you.

[jonforums (Jon Forums)]

No failure when building trunk@37710 on Arch 3.6.6 with gcc 4.7.2

[h.shirosaki (Hiroshi Shirosaki)]

This seems due to r37337. I don't know the reason of r37337, but reverting the commit (use rb_w32_fstati64 for fstat) will fix this.

[usa (Usaku NAKAMURA)]

• Assignee changed from usa (Usaku NAKAMURA) to nobu (Nobuyoshi Nakada)

[jonforums (Jon Forums)]

This still fails as of r38131 on my Win7 32bit system. Given Shirosaki-san's comment, what's preventing this from being quickly fixed on trunk?

Since this bug is not made visible by the rubyinstaller CI (build slave location) I will manually run any tests or builds you need in order to validate the fix. Let me know what you need.

[jonforums (Jon Forums)]

Status? Is this planned to be fixed for 2.0.0?

This is the only failure I get when building trunk (make test-all and make test) with mingw-w64 4.7.2 32bit on Win7.

As stated previously, you don't see this failure at http://ci.rubyinstaller.org/ because Luis' build slaves are located in an uneffected timezone.

[phasis68 (Heesob Park)]

This daylight savings time bug is described at
http://www.codeproject.com/Articles/1144/Beating-the-Daylight-Savings-Time-bug-and-getting
and
http://support.microsoft.com/kb/158588/en-us

The workaround is using the Windows API call instead of using the C library call.

[jonforums (Jon Forums)]

Nobu-san: why was r37337 required and rb_w32_fstati64 no longer correct?

Heesob: good read. After reverting r37337 as per Hiroshi's comment, I get 0 fails on trunk@38966 (Win7 32bit + mingw-w64 4.7.2) because fstat -> rb_w32_fstati64 (win32/win32.c:4474) rather than fstat -> _fstati64 which causes mingw-w64 to use msvcrt's _fstat64 (i686-w64-mingw32/include/_mingw_stat64.h:12)

http://msdn.microsoft.com/en-us/library/221w8e43%28v=vs.90%29.aspx

Given the info you linked, do you think the current rb_w32_fstati64 is a valid workaround? Does reverting r37337 also work for you on Win8?

[phasis68 (Heesob Park)]

Reverting r37337 works on Windows 8.

I think the only drawback to rb_w32_fstati64 is a little performance loss.

[nobu (Nobuyoshi Nakada)]

I don't remember the reason well, probably some mingw failed to link the dll or extension libraries.
But I can't find which failed, so revert it.

[nobu (Nobuyoshi Nakada)]

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

This issue was solved with changeset r38992.
Jon, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

win32.h: revert r37337

• include/ruby/win32.h (fstat): revert r37337, which uses _fstati64()
  instead of fstati64() on mingw32. [Bug #7276]