[omniORB] giopServer::deactivate hangs

[baileyk at schneider.com]

duncan at grisby.org wrote:
> Forking from multi-threaded programs often causes odd things to
> happen. It's safer to Pre-fork before B becomes multi-threaded, then
> have a single threaded forker, F, which talks to B through pipes. That
> way, C would be a child of F, rather than directly of B.

The only thing I do between fork and exec is dup2() and close() calls.
Hopefully I'm OK.  My Solaris app is using libpthread, so fork() is acting
like the Solaris fork1().

> It's not guaranteed to be safe to signal a condition variable from
> inside a signal handler. The only threading primitive which is
> guaranteed safe is the System V semaphore.

That makes sense.  You specifically say "System V semaphore", but I've done
some reading and it appears the POSIX sem_post is async-signal safe.  The
Linux sem_post man page says

DESCRIPTION
       This  manual  page documents POSIX 1003.1b semaphores, not
       to be confused with SystemV  semaphores  as  described  in
       ipc(5), semctl(2) and semop(2).

and later

ASYNC-SIGNAL SAFETY
       On processors supporting  atomic  compare-and-swap  (Intel
       486,  Pentium and later, Alpha, PowerPC, MIPS II, Motorola
       68k), the sem_post function is async-signal safe  and  can
       therefore be called from signal handlers. This is the only
       thread synchronization function provided by POSIX  threads
       that is async-signal safe.

       On  the  Intel 386 and the Sparc, the current LinuxThreads
       implementation of sem_post is  not  async-signal  safe  by
       lack of the required atomic operations.

The Solaris sem_post man page claims async-signal safety though.  So I can
use sem_post on Solaris/Sparc or Linux/Intel, but not Linux/Sparc?  I'm OK
with that.  I'll use a POSIX semaphore to wake the thread that calls
ORB::shutdown(), or maybe just use sigwait().  It would be nice if omniORB
provided an API for "shutdown by signal" even though there's no OMG
standard for one.  I'd prefer not to use a CORBA invocation to trigger
shutdown.

The hanging threads weren't related to the condition variable though.  I'll
try a CVS omniORB snapshot sometime.  Right now I can get by without this
stuff working perfectly.  I'll post results if I get around to rebuilding a
omniORB.

Thanks,
Kendall