[Boost-users] [Threads] join_all() strange behavior?

From: Manuel Jung (gzahl_at_[hidden])
Date: 2007-01-10 06:21:52

Hi,

I have a strange problem. The destructor of a class calls join_all().
But it seems that he hangs there. Even though some other code behind the
join_all() is executed before it finishs! 
Lets get concret:

seal::~seal()
{
ThgPrgs.join_all();
cout<<"Crabd quits, goodbye"<<endl;
}

When i run my program, the last message is "Crabd quits, goodbye" and it
should exit then, but it doesnt. If i stop it running in gdb an do a
backtrace i get the following:

#0 Â 0xffffe410 in __kernel_vsyscall ()
#1 Â 0xb7c796c8 in pthread_join () from /lib/tls/i686/cmov/libpthread.so.0
#2 Â 0xb7d21e6a in boost::thread::join ()
from /usr/local/lib/libboost_thread-gcc-mt-1_33_1.so.1.33.1
#3 Â 0xb7d224ee in boost::thread_group::join_all ()
   from /usr/local/lib/libboost_thread-gcc-mt-1_33_1.so.1.33.1
#4 Â 0x0808742c in ~seal (this=0x80c9890) at src/seal.cpp:20
#5 Â 0x0808eb55 in checked_delete<seal> (x=0x80c9890)
at /usr/local/include/boost/checked_delete.hpp:34
#6 Â 0x0809566a in boost::detail::sp_counted_impl_p<seal>::dispose
(this=0x80dc410)
    at /usr/local/include/boost/detail/sp_counted_impl.hpp:76
#7 Â 0x0805950f in boost::detail::sp_counted_base::release (this=0x80dc410)
    at /usr/local/include/boost/detail/sp_counted_base_gcc_x86.hpp:145
#8 Â 0x08059555 in ~shared_count (this=0x8103e04)
at /usr/local/include/boost/detail/shared_count.hpp:159
#9 Â 0x0808dd1c in ~shared_ptr (this=0x8103e00) at src/seal.h:39
#10 0x08089182 in ~tcp_connection (this=0x8103df8) at src/seal.cpp:212
#11 0x08094273 in checked_delete<tcp_connection> (x=0x8103df8)
    at /usr/local/include/boost/checked_delete.hpp:34
#12 0x080967d0 in boost::detail::sp_counted_impl_p<tcp_connection>::dispose
(this=0x80dcf00)
    at /usr/local/include/boost/detail/sp_counted_impl.hpp:76
#13 0x0805950f in boost::detail::sp_counted_base::release (this=0x80dcf00)
    at /usr/local/include/boost/detail/sp_counted_base_gcc_x86.hpp:145
#14 0x08059555 in ~shared_count (this=0x80dc4e8)
at /usr/local/include/boost/detail/shared_count.hpp:159
#15 0x0808ddc6 in ~shared_ptr (this=0x80dc4e4) at src/seal.h:66
#16 0x0808e69f in ~value (this=0x80dc4e4) at src/seal.cpp:232
#17 0x0808e700 in ~list3 (this=0x80dc4e0) at src/seal.cpp:232
#18 0x0808e766 in ~bind_t (this=0x80dc4d8) at src/seal.cpp:232
#19 0x080910a7 in ~accept_handler (this=0x80dc4c0)
    at /usr/local/include/boost/asio/detail/reactive_socket_service.hpp:1228
---Type <return> to continue, or q <return> to quit---
#20 0x08095595 in ~op (this=0x80dc4b0)
at /usr/local/include/boost/asio/detail/reactor_op_queue.hpp:355
#21 0x080918db in
boost::asio::detail::reactor_op_queue<int>::op
<boost::asio::detail::reactive_socket_service<boost::asio::ip::tcp,
boost::asio::detail::epoll_reactor<false>
>::accept_handler<boost::asio::basic_socket<boost::asio::ip::tcp,
boost::asio::stream_socket_service<boost::asio::ip::tcp> >,
boost::_bi::bind_t<void, boost::_mfi::mf2<void, tcp_server,
boost::shared_ptr<tcp_connection>, boost::asio::error const&>,
boost::_bi::list3<boost::_bi::value<tcp_server*>,
boost::_bi::value<boost::shared_ptr<tcp_connection> >, boost::arg<1> > > >
>::destroy_handler (base=0x80dc4b0)
at /usr/local/include/boost/asio/detail/reactor_op_queue.hpp:355
#22 0x080709d2 in
boost::asio::detail::reactor_op_queue<int>::op_base::destroy
(this=0x80dc4b0)
    at /usr/local/include/boost/asio/detail/reactor_op_queue.hpp:294
#23 0x0806df75 in
boost::asio::detail::reactor_op_queue<int>::destroy_operations
(this=0x80c99a8)
    at /usr/local/include/boost/asio/detail/reactor_op_queue.hpp:245
#24 0x0806d677 in
boost::asio::detail::epoll_reactor<false>::shutdown_service
(this=0x80c9970)
    at /usr/local/include/boost/asio/detail/epoll_reactor.hpp:108
#25 0x0806c2ec in ~service_registry (this=0x80c98d8)
    at /usr/local/include/boost/asio/detail/service_registry.hpp:53
#26 0x0806ba81 in ~io_service (this=0x80c98d4) at src/seal.cpp:8
#27 0x08087648 in ~seal (this=0x80c9890) at src/seal.cpp:32
#28 0x0808eb55 in checked_delete<seal> (x=0x80c9890)
at /usr/local/include/boost/checked_delete.hpp:34
#29 0x0809566a in boost::detail::sp_counted_impl_p<seal>::dispose
(this=0x80da710)
    at /usr/local/include/boost/detail/sp_counted_impl.hpp:76
#30 0x0805950f in boost::detail::sp_counted_base::release (this=0x80da710)
    at /usr/local/include/boost/detail/sp_counted_base_gcc_x86.hpp:145
#31 0x08059555 in ~shared_count (this=0xbfa2e124)
at /usr/local/include/boost/detail/shared_count.hpp:159
#32 0x0808dd1c in ~shared_ptr (this=0xbfa2e120) at src/seal.h:39
#33 0x080a8d68 in boost::shared_ptr<seal>::reset (this=0x80c8824)
    at /usr/local/include/boost/shared_ptr.hpp:231
#34 0x080a863e in main (argc=1, argv=0xbfa2e204) at src/crabd.cpp:71

The class seal holds some other classes in shared_ptrs which are will be
destroyed in this moment too. Some of them have also a join_all() in their
destructors. Im wondering if they produce the lock?

But the main question here is:
I read "Crabd quits, goodbye" and after this i stop the program and i get in
the backtrace from gdb the line "#4 Â 0x0808742c in ~seal (this=0x80c9890)
at src/seal.cpp:20". And Line 20 in seal.cpp is the line with join_all().
How is this possible, though join_all should block?

Thanks for your help and for reading this.
Manuel Jung, Germany