From: Dane Jensen <dane.jensen@...>
Date: 2010-02-21T15:37:12+09:00
Subject: Re: segfaults from the 'object-free' dtrace probe

On Sat, Feb 20, 2010 at 4:23 PM, Dane Jensen <dane.jensen@gmail.com> wrote:
>
> Hi All-
>
> Has anyone had any luck with the 'object-free' dtrace probe... ever... on
> any platform? It segfaults ruby without fail, absolutely 100% of the time,
> everywhere.
>
> It did it in Leopard, and there's a thread about it here:
> http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/287796 .
>
> I tried testing it again yesterday in Snow Leopard and got the same
> segfault. Then in OpenSolaris 2009.06, and then OpenSolaris b133 that came
> out last night. Without fail they all die after the probe fires.
>
> There was a patch mentioned in the thread above, did that ever come about?
>
> -Dane

Ahhh... I think I see what it's doing, at least in a simple test case
I just cooked up that doesn't segfault, but rather explodes with:
[BUG] object allocation during garbage collection phase

The object-free probe fires during GC in obj_free. The entire probe is
two lines, most of which is dtrace boilerplate.. So the only thing
that could be causing it is:
    rb_class2name(CLASS_OF(obj)));

So... that ends up creating an object in there I assume, right?

-Dane