Re: ruby1.9.1 hangs again

[Petr Salinger <Petr.Salinger@seznam.cz>]

Hi.

> A change in Ruby caused it to hang again, likely due to a pthread
> semantic difference on kfreebsd. It doesn't hang, neither on Linux nor
> on FreeBSD.
>
> Affected package is the experimental one. It's a pre-release snapshot
> (candidate RC1) for Ruby 1.9.3, to be released in september, and likely
> the default Ruby implementation in wheezy.
>
> Please see http://redmine.ruby-lang.org/issues/5240

Many thanks for bisecting it.
I looked at the diff 32243:32244.

The change in general should work even for our slightly broken pthread semantic.
It should be possible to workaround it by encapsulating 
timer_thread_pipe_owner_proces usage.

The idea is:

standard code would be

***
bool timer_thread_running()
{
   return timer_thread_pipe_owner_process == getpid();
}

void timer_thread_set()
{
   timer_thread_pipe_owner_process = getpid();
}

void timer_thread_child_at_fork()
{
}
***

and somewhere during initialization

pthread_atfork(NULL, NULL, timer_thread_child_at_fork)

And our code will be

***
bool timer_thread_running()
{
   return timer_thread_pipe_owner_process == 1;
}

void timer_thread_set()
{
   timer_thread_pipe_owner_process = 1;
}

void timer_thread_child_at_fork()
{
   timer_thread_pipe_owner_process = 0;
}
***

Even on fully conforming implementation might be our variant better,
as getpid() might be costly system call.

Thanks again for investigating it.

	Petr