Re: __cxa_atexit and "FAIL: cxxtest run"

[Ranjit Mathew <rmathew at gmail dot com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[Please CC me on your replies. Thanks.]

Benjamin Kosnik wrote:
>>FWIW, this looks somewhat similar to PR libstdc++/23591:
>>
>>  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23591
>>
>>It looks like libstdc++.so.6 was unloaded a bit too
>>prematurely for the exit handlers and xdr_u_long_internal()
>>is just a red herring for the real problem.
> 
> 
> This particular bug is fixed in mainline. 
> 
> Your analysis still seems pretty hazy to me.

Because it's just a hypothesis and PR23591 looks
somewhat similar. (The other thread is waiting
on a pthread_cond_wait() in my case when the first thread
dies in exit().) I knew that PR23591 was fixed on mainline
but the symptoms look similar.

I briefly looked at the glibc source code corresponding
to the installed glibc on the box and I can't see how
xdr_u_long_internal() (or xdr_u_int() which calls it)
gets called in any exit handler. Hence the suspicion
that it's a red herring and a wrong mapping of virtual
address to a library symbol.

On my home PC, where I have a vanilla FSF glibc-2.3.5
that's compiled NPTL-only, I cannot reproduce this
problem, while on another PC where I have RHEL U5's
glibc-2.3.2, I still can reproduce the problem. So
there's either a problem with glibc-2.3.2 (+RHEL's
changes) or for some other reason the problem is only
manifesting itself with one setup and not the other.

If it's a glibc bug that has now been fixed, I'm not
too concerned and would not spend time in wasting the
time of the people on these lists.

Thanks,
Ranjit.

- --
Ranjit Mathew      Email: rmathew AT gmail DOT com

Bangalore, INDIA.    Web: http://ranjitmathew.hostingzero.com/


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (MingW32)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDpn6iYb1hx2wRS48RAiceAJ0dKmfGwv+EseJCbV+zE/h8wiN5qgCgm05k
0dY66r5jH/+aMiCmF+R0Dhg=
=z3qP
-----END PGP SIGNATURE-----