Bug #7109: File.utime doesn't set nanoseconds

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: ruby 1.9.3p194 (2012-04-20 revision 35410) [x86_64-linux]

Hi,
I'm having a problem with File.utime on RHEL 5/CentOS 5. The File.utime method seems not to set nanoseconds properly, see the attached test case, that fails (the nanoseconds get rounded to thousand, so in fact I get microseconds). The test doesn't seem to fail on newer platforms with the same Ruby version, though (RHEL 6/CentOS 6, Fedora 17). Note, that when creating/modifying files, the nanoseconds get set and are returned properly, so it seems that they are supported by filesystem.

Sample failure:

1. Failure:
   test_utime(TestUtime) [/builddir/build/BUILD/rubygem-sprockets-2.4.5/opt/rh/ruby193/root/usr/share/gems/gems/sprockets-2.4.5/file_utime_fail.rb:10]:
   Expected: 391199303
   Actual: 391199000

Any idea what might be wrong?
Thanks!

[akr (Akira Tanaka)]

2012/10/5 bkabrda (Bohuslav Kabrda) [e-mail]:

I'm having a problem with File.utime on RHEL 5/CentOS 5. The File.utime method seems not to set nanoseconds properly, see the attached test case, that fails (the nanoseconds get rounded to thousand, so in fact I get microseconds). The test doesn't seem to fail on newer platforms with the same Ruby version, though (RHEL 6/CentOS 6, Fedora 17). Note, that when creating/modifying files, the nanoseconds get set and are returned properly, so it seems that they are supported by filesystem.

utimensat system call is required to set nanosecond filestamp.
It is available since Linux 2.6.22.
(glibc support is also required but I'm not sure the exact version.)

I guess RHEL 5/CentOS 5 is older than that.

Tanaka Akira

----- Original Message -----

2012/10/5 bkabrda (Bohuslav Kabrda) [e-mail]:

I'm having a problem with File.utime on RHEL 5/CentOS 5. The
File.utime method seems not to set nanoseconds properly, see the
attached test case, that fails (the nanoseconds get rounded to
thousand, so in fact I get microseconds). The test doesn't seem to
fail on newer platforms with the same Ruby version, though (RHEL
6/CentOS 6, Fedora 17). Note, that when creating/modifying files,
the nanoseconds get set and are returned properly, so it seems
that they are supported by filesystem.

utimensat system call is required to set nanosecond filestamp.
It is available since Linux 2.6.22.
(glibc support is also required but I'm not sure the exact version.)

I guess RHEL 5/CentOS 5 is older than that.

Tanaka Akira

True, kernel version is 2.6.19. So is there another way to solve this or do I have to get along with microseconds?

Thanks.

--
Regards,
Bohuslav "Slavek" Kabrda.

[akr (Akira Tanaka)]

2012/10/5 Bohuslav Kabrda [e-mail]:

True, kernel version is 2.6.19. So is there another way to solve this or do I have to get along with microseconds?

We can't set a timestamp without appropriate system call.

Tanaka Akira

[kosaki (Motohiro KOSAKI)]

utimensat system call is required to set nanosecond filestamp.
It is available since Linux 2.6.22.
(glibc support is also required but I'm not sure the exact version.)

I guess RHEL 5/CentOS 5 is older than that.

True, kernel version is 2.6.19. So is there another way to solve this or do I have to get along with microseconds?

No another way. This is kernel limitation. In other word, you can't
compare gettimeofday() result and stat() result straightforwardly.

btw, even tough you use RHEL6, you can see this issue if you use some
older filesystems.
it depend on filesystem.

btw2, http://en.wikipedia.org/wiki/Ext4 says nanosecond timestamp is a
new feature of ext4.
I suspect your "RHEL5" mean ext3.

[mame (Yusuke Endoh)]

• Status changed from Open to Rejected

It cannot be avoided because the limitation is due to OS. Closing.

--
Yusuke Endoh [e-mail]