From: Nobuyoshi Nakada <nobu@...>
Date: 2007-09-23T09:20:20+09:00
Subject: Re: Next 1.8.6 on Sept. 22

Hi,

At Sun, 23 Sep 2007 01:01:48 +0900,
Sylvain Joyeux wrote in [ruby-core:12226]:
> I completely agree on that. assert_no_survivors() is IMO only triggered 
> when a thread is forcefully killed (i.e. thread killed without calling 
> ensure blocks).

Do you consider the assertions would fail even when it isn't
fatal condition?  In other words, the failure doesn't mean a
bug of core and/or ext/thread?

> Two things worry me though ...
> 1/ the warning that some threads are still running when the mutex is killed 
> is definitely useful one way or the other. I just had a test passing while 
> there were a deadlock somewhere. The deadlock has been notified to me by 
> that warning ...

It wasn't a warning, but was dumping the core.  If it could
occur in non-fatal condition, it must not use rb_bug(), but
should raise an exception instead, as the current code does.

> 2/ nothing of that stuff has been committed to 1.8.6. Will we have another 
> patch release with a broken threading support ? That worries me more than 
> the first point ...

Seems it has been backported this morning (in JST).

-- 
Nobu Nakada