[ruby-core:39620] [Ruby 1.9 - Bug #5335] [RFC/PATCH] test_old_thread_select: timing tweaks

Issue #5335 has been updated by Motohiro KOSAKI.

Assignee changed from Motohiro KOSAKI to Eric Wong

> diff --git a/test/-ext-/old_thread_select/test_old_thread_select.rb b/test/-ext-/old_thread_select/test_old_thread_select.rb
> index 1ccdb34..a8a0ce2 100644
> --- a/test/-ext-/old_thread_select/test_old_thread_select.rb
> +++ b/test/-ext-/old_thread_select/test_old_thread_select.rb
> @@ -16,10 +16,10 @@ class TestOldThreadSelect < Test::Unit::TestCase
>    def test_old_select_read_timeout
>      with_pipe do |r, w|
>        t0 = Time.now
> -      rc = IO.old_thread_select([r.fileno], nil, nil, 0.001)
> +      rc = IO.old_thread_select([r.fileno], nil, nil, 0.01)
>       diff = Time.now - t0
>        assert_equal 0, rc
> -      assert diff >= 0.001, "returned too early"
> +      assert_operator diff, :>=, 0.004, "returned too early: diff=#{diff}"

No, I really dislike this. both 0.01 and 0.004 make no sense and no good reason we choose them.
I guess 0.004 mean 250Hz. But in fact, only several linux systems use 250Hz. No platform independent meanings.

Instead, I just recommend to skip this test if platform is too old linux.
(I guess our chkbuild caught the same issue, http://www.rubyist.net/~akr/chkbuild/debian/ruby-trunk/log/20110919T060500Z.log.html.gz)


In addition, the correct way is, to detect inaccurate select timeout by configure script and do_select() retry to call select(2)
automatically _if_ we really need to care.
 

----------------------------------------
Bug #5335: [RFC/PATCH] test_old_thread_select: timing tweaks
http://redmine.ruby-lang.org/issues/5335

Author: Eric Wong
Status: Open
Priority: Normal
Assignee: Eric Wong
Category: core
Target version: 1.9.3
ruby -v: ruby 1.9.3dev (2011-09-17 revision 33290) [x86_64-linux]


I was getting timing errors and short select() timeouts on CentOS 5.6
2.6.18-238.9.1.el5xen most likely due to CONFIG_HZ=250 and the lack of dynticks,
but maybe being a Xen VM has this effect, too.

select() itself appears to just be inaccurate on this system:

	strace -e select -T ./ruby -e 'select([STDIN],nil,nil,0.001)'
	select(1, [0], NULL, NULL, {0, 1000})   = 0 (Timeout) <0.000046>

I'm not happy with timing tests, but I can't think of another way to test
functionality like this.


-- 
http://redmine.ruby-lang.org