Re: kgio test_signal_close failed

[Eric Wong <normalperson@yhbt.net>]

375gnu <375gnu@gmail.com> wrote:
> On 3/4/12, Eric Wong <normalperson@yhbt.net> wrote:
> > 375gnu <375gnu@gmail.com> wrote:
> >> Sometimes test_signal_close from test_poll.rb fails. Not every time, and
> >> only
> >> with ruby1.9 (1.9.3.0 is my current version). With ruby 1.8.7 it's passed.
> > Is this also on Debian/kfreebsd?
> 
> Not (0 of 25 times). But it runs in kvm, so it's much slower than real linux
> box (110 sec for all tests vs. 75).

Odd, I can't reproduce under kvm (Debian wheezy + custom 3.3.0-rc4
kernel).  Are you on 2.6.38 kernel by any chance?  Some versions of that
had weird signal issues.

Also, this test is just skipped under Matz Ruby 1.8 (can't safely
support poll() w/o native threads + GVL release).

> >>     59:     thr = Thread.new { sleep 0.100; Process.kill(:USR1, $$) }
> >
> > Does increasing the sleep time there help?  It's a potential race
> > condition :/
> 
> No, even with "sleep 2" (1 of 10 is failed)

Perhaps this makes the test less bad?

diff --git a/test/test_poll.rb b/test/test_poll.rb
index ba0e561..df59354 100644
--- a/test/test_poll.rb
+++ b/test/test_poll.rb
@@ -58,7 +58,10 @@ class TestPoll < Test::Unit::TestCase
     res = nil
     thr = Thread.new { sleep 0.100; Process.kill(:USR1, $$) }
     t0 = Time.now
-    assert_raises(IOError) { Kgio.poll({@rd => Kgio::POLLIN}) }
+    assert_raises(IOError) do
+      result = Kgio.poll({@rd => Kgio::POLLIN})
+      result.each_key { |io| io.read_nonblock(1) }
+    end
     diff = Time.now - t0
     thr.join
     assert diff >= 0.010, "diff=#{diff}"