Re: [Boost-users] pthread_exit() from boost::thread segfaults

Subject: Re: [Boost-users] pthread_exit() from boost::thread segfaults
From: Viatcheslav.Sysoltsev_at_[hidden]
Date: 2010-12-21 04:55:04

>>
>> There is no reason for your snippet to fail. I have tested on openSUSE  
>> 11.3,
>> 64 bits with gcc 4.5.0 and libboost_thread 1.42.
>> Try to remove from the thread starter function the call to  
>> pthread_exit() and
>> see if you still get the error.
> It works well without pthread_exit(), I'm using boost about an year on  
> this machine unchanged, that's why this fault wonder me sooo much.
>
> update to the issue: the prog freezes with latest boost release 1.45 in  
> the main thread, the child thread exits without any error:
>
> (LD_LIBRARY_PATH is set to /home/vsysolts/boost/boost_1_45_0/stage/lib)
>
>
> #include <stdio.h>
> #include <boost/thread.hpp>
>
> void tf()
> {
>      printf("thread\n");
>      sleep(1);
>      pthread_exit(NULL);
> }
>
>
> int main()
> {
>      printf("main entered\n");
>      boost::thread t(tf);
>      t.join();
>      printf("main exiting");
>      return  0;
> }
>
>
> vsysolts:~/test/boost_thread (0) > g++  
> -I/home/vsysolts/boost/boost_1_45_0 test2.cpp  
> -L/home/vsysolts/boost/boost_1_45_0/stage/lib -lboost_thread && gdb a.out
> GNU gdb (GDB) 7.2
> Copyright (C) 2010 Free Software Foundation, Inc.
> License GPLv3+: GNU GPL version 3 or later  
> <http://gnu.org/licenses/gpl.html>
> This is free software: you are free to change and redistribute it.
> There is NO WARRANTY, to the extent permitted by law.  Type "show  
> copying"
> and "show warranty" for details.
> This GDB was configured as "x86_64-unknown-linux-gnu".
> For bug reporting instructions, please see:
> <http://www.gnu.org/software/gdb/bugs/>...
> Reading symbols from /home/vsysolts/test/boost_thread/a.out...done.
> (gdb) r
> Starting program: /home/vsysolts/test/boost_thread/a.out
> [Thread debugging using libthread_db enabled]
> main entered
> [New Thread 0x7ffff6cce950 (LWP 12763)]
> thread
> [Thread 0x7ffff6cce950 (LWP 12763) exited]
> ^C
> Program received signal SIGINT, Interrupt.
> 0x00007ffff6cd9d59 in pthread_cond_wait@@GLIBC_2.3.2 () from  
> /lib64/libpthread.so.0
> (gdb) bt
> #0  0x00007ffff6cd9d59 in pthread_cond_wait@@GLIBC_2.3.2 () from  
> /lib64/libpthread.so.0
> #1  0x00007ffff7bd6184 in boost::thread::join() () from  
> /home/vsysolts/boost/boost_1_45_0/stage/lib/libboost_thread.so.1.45.0
> #2  0x0000000000402d45 in main ()
>
> -- Slava

Hmm, looking into thread.cpp implementation of thread_proxy (callback  
function for pthread_create()) and boost::thread::join I do not see how  
this join implementation would work with any abnormal thread termination.
In thread.cpp boost::thread::join waits the thread termination on line:
local_thread_info->done_condition.wait(lock);

The lock gets signaled only at end of thread_proxy(), which is reached  
only if the user thread routine does "return" or is interrupted with  
boost::thread means. pthread_exit(), as well as any other emergency  
termination does only stack unwinding (on most but not all systems), so  
done_condition is simply never signaled in these circumstances. To be  
correct the thread_proxy() should have installed cleanup routine with  
pthread_cleanup_push() and, to be sure, have an local object which  
destructor would signal done_condition as well, but I do not see it in  
boost::thread code.

Bogdan, when you said, the test program works for you - did you means it  
does return with status 0? If so, I wonder, which way the execution goes  
in your thread_proxy implementation - can you please debug it a little if  
you have some time? To build debug version of boost you need to add  
"variant=debug" option to bjam. I tried the same test program on ubuntu  
10.04 TLS with gcc 4.4.3 and boost 1.40 - the same result as with latest  
boost: it freezes in boost::thread::join() on condition.wait().

-- Slava