Re: [Boost-users] [Serialization]: SIGSEGV on Mingw32, during global static data initialization (runs ok on Linux)

From: Jeffrey Chang (jeffrey8chang_at_[hidden])
Date: 2006-09-27 14:49:02

Robert:

Thanks for the info.  From the gdb output, my program was still
single-threaded when it seg-faults:

(gdb) info thread
* 1 thread 3800.0x3f8  0x6e3759a3 in
boost::archive::detail::basic_serializer::basic_serializer (this=0x4,
eti=@0x816484)
    at
c:/e2e/prj/awit.hipeer.trunk/contrib/oss.boost.release/boost_1_33_1/boost/archive/detail/basic_serializer.hpp:37

I do find a workaround for my problem:  use a single complication unit (*.o)
for all my *.cpp files that does #include serialization headers, which are
<boost/serialization/nvp.hpp>, <boost/serialization/shared_ptr.hpp>, and
<boost/serialization/vector.hpp>.  These *.cpp files used to scatter in
multiple *.o files, which are archived into a lib*.a file before statically
linked with the one with main().

Since the program halts at the get_instance() call at line
boost_1_33_1/boost/serialization/extended_type_info_typeid.hpp:95, I'm
wondering the discussion a few months ago: Re: Boost serialization :
concurrency (was
Possibletoappend...)<http://news.gmane.org/find-root.php?message_id=%3ceektei%248hm%241%40sea.gmane.org%3e>
has anything to do with the safety of local static access across multiple
C++ compilation unit, in addition to issues with thread-safety.  Anyway, is
there any documentation on how to use boost::serialization library correctly
across multiple DLLs & explain what it is doing during
initialization/destruction time?

Again, thanks for your hint, as well as the wonderful boost serialization
library.

--- Jeffrey

On 9/26/06, Robert Ramey <ramey_at_[hidden]> wrote:
>
>  It has been recently discovered that, contrary to intention and belief,
> archives are not currently thread-safe.
>
> The issue is currently being investigated but a fix won't appear before
> version 1.35 - at the earliest.
>
> In the mean time, perhaps you can avoid the issue by sycronizing access to
> archives.
>
> Robert Ramey
>
> "Jeffrey Chang" <jeffrey8chang_at_[hidden]> wrote in message
> news:8782a5590609261636j3ee55435pe578c78290ec7d1c_at_mail.gmail.com...
> Robert:
>
> Yes, this program uses boost multi-thread library of
> "boost_thread-mgw-mt-d-1_33_1.dll".  boost::asio library (header only) is
> also used.
>
> --- Jeffrey
>
> On 9/26/06, Robert Ramey <ramey_at_[hidden]> wrote:
> >
> >  Just out of curiosity? are you doing multi-threading?
> >
> > Robert Ramey
> >
> > "Jeffrey Chang" <jeffrey8chang_at_[hidden]> wrote in message news:8782a5590609261453p6f8542cfp2cb64d86754527c7_at_[hidden]
> > ...
> > Hi,
> >
> > I've been using boost Serialization library (release v1_33_1) on Linux
> > without problems, with g++ compiler of both v3.3.6 & v4.1.1.  My program
> > serialzes boost::shared_ptr of polymorphic types, and BOOST_CLASS_EXPORT()
> > and BOOST_SERIALIZATION_SHARED_PTR() are used.
> >
> > However, when the same program is ported & run on mingw32 (w/ g++ v3.4.5on WinXP), it generates SIGSEGV before my main() is called.  and here's the
> > stack trace:
> >
> > Program received signal SIGSEGV, Segmentation fault.
> > 0x6e3759a3 in boost::archive::detail::basic_serializer::basic_serializer
> > (
> >     this=0x4, eti=@0x813484)
> >     at
> > /opt/boost/boost_1_33_1/boost/archive/detail/basic_serializer.hpp:37
> > 37          {}
> > (gdb) bt
> > #0  0x6e3759a3 in
> > boost::archive::detail::basic_serializer::basic_serializer (
> >     this=0x4, eti=@0x813484)
> >     at
> > /opt/boost/boost_1_33_1/boost/archive/detail/basic_serializer.hpp:37
> > #1  0x6e34220b in
> > boost::archive::detail::basic_oserializer::basic_oserializer
> >     (this=0x0, eti=@0x813484)
> >     at
> > /opt/boost/boost_1_33_1/libs/serialization/build/../src/basic_oserializer.cpp:24
> > #2  0x005fb394 in boost::archive::detail::pointer_oserializer<HmcBase,
> > boost::archive::text_oarchive>::pointer_oserializer (this=0x813164)
> >     at
> > /opt/boost/boost_1_33_1/boost/serialization/extended_type_info_typeid.hpp:95
> > #3  0x00477c12 in __static_initialization_and_destruction_1 (
> >     __initialize_p=8467588, __priority=65535)
> >     at /opt/boost/boost_1_33_1/boost/archive/detail/oserializer.hpp:198
> > #4  0x004c5a57 in __main ()
> >     at /opt/boost/boost_1_33_1/boost/archive/detail/oserializer.hpp:198
> > #5  0x00404568 in main (argc=1, argv=0x3d45f8) at
> > ../../src/HmcMain.cpp:231
> > (gdb)
> >
> >
> > I cannot connect the logic between stack frame #3 and #2; it seems
> > trying to instantiate some global static data when SIGSEGV happens.  Is
> > there some dependency on order of global static data initialization  to be
> > satisfied for using BOOST_CLASS_EXPORT/BOOST_SERIALIZATION_SHARED_PTR?
> > HmcBase is the base class for my polymorphic types, and all my code are
> > statically linked, while "boost_serialization-mgw-d-1_33_1.dll" is used
> > here.    I saw some discussion in this mailing list on "Serializing
> > classes defined in DLLs"
> > <http://news.gmane.org/find-root.php?message_id=%3c432FF0C3.5010002%40intopii.fi%3e>a
> > year ago, but don't know what's the current status on it and whether this
> > case is related to it.
> >
> > Thanks for any hit,
> >
> > --- Jeffrey
> >
> > ------------------------------
> >
> > _______________________________________________
> > Boost-users mailing list
> > Boost-users_at_[hidden]
> > http://lists.boost.org/mailman/listinfo.cgi/boost-users
> >
> >
> > _______________________________________________
> > Boost-users mailing list
> > Boost-users_at_[hidden]
> > http://lists.boost.org/mailman/listinfo.cgi/boost-users
> >
> >
>  ------------------------------
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>
>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>
>