From: Charles O Nutter <headius@...>
Date: 2004-10-31T12:44:27+09:00
Subject: Possible flaw in rubicon test for TestThreadGroup

Hello! I'm trying to implement Thread and ThreadGroup semantics for
JRuby, and have run into an issue with one of the rubicon tests;
specifically, the "test_list" test in TestThreadGroup.

The test is as follows:
  def test_list
    tg = ThreadGroup.new
    10.times do
      Thread.critical = true
      t = Thread.new { Thread.stop }
      tg.add(t)
    end
    assert_equal(10, tg.list.length)
    tg.list.each {|t| t.wakeup; t.join }
  end

The problem I'm running into is that there's no guarantee that
Thread.stop will be called in the ten threads created before t.wakeup
is called below. In JRuby, since Ruby threads are mapped directly to
JVM threads (and by extension, to native threads), this test usually
just freezes up. That's because t.wakeup is called before the thread
has been scheduled to run; t.join is then called, the thread in
question Thread.stops, and that's that.

If I replace the tg.list.each... line above with the following:
    tg.list.each do |t| 
      Thread.pass while (t.status != "sleep")
      t.wakeup
      t.join
    end

...The test succeeds every time. The Thread.pass line is also very
common throughout tests in TestThread.rb.

I'm trying to determine whether this is a flaw in the way I've
implemented stop and wakeup, or if there's an issue with this test.
Comments?