Re: problem with pthread and waitpid()

[Britton <fsblk at aurora dot uaf dot edu>]

> > I have noticed a problem with the LinuxThreads implementation.  When a
> > process forks and execs a program that spawns threads, then waitpids on
> > that child, the wait ends when the initial thread of the exec'd program
> > finishes, but before the spawned threads are necesarilly dead.  
> 
> Which version of LinuxThreads and how are the threads terminating?

I guess I skipped the important part of the story, which is that the
program getting waitpid()ed on is getting SIGTERM ed first.  So far as I
can tell everything works correctly when there is no such fatal signal
delivered to the initial thread.  But this (from libc/linuxthreads/README
is not strictly true

  2- A fatal asynchronous signal terminates all threads in the process.
     (OK. The thread manager notices when a thread dies on a signal
      and kills all other threads with the same signal.)

There is a timing glitch during which the initial thread may be dead but
resource hungry children still alive.  I guess this falls into the
category of things that can't be fixed until CLONE_PID works.  Anyone know
if it does in 2.4?

> I addressed a couple of bugs related to LinuxThreads shutdown since 2.1.3. The
> two that I recall off the top of my head are:
> 
> - a potential unnecessary two second delay in shutdown due to the 2000
>   millisecond poll() timeout in the thread manager's loop.

Thank you thank you thank you.  I have already been bitten by this.


Britton