Re: [boost] [Boost.Serialization] Crash in current master (1.68)

Subject: Re: [boost] [Boost.Serialization] Crash in current master (1.68)
From: Robert Ramey (ramey_at_[hidden])
Date: 2018-06-29 15:20:47

On 6/29/18 2:07 AM, Alexander Grund via Boost wrote:
> Hi all again,
> 
> I have to stress the importance of a swift action/review here as the 
> current master now contains the flawed commit again that leads to a 
> certain crash when using shared libraries linked against static boost.

Right now I'm bogged down in other stuff, but I did find a little time 
to investigate this some more.  I reviewed the information on the 
failures if test_dll_export along with the corresponding config.info 
output.  I've found some useful things.  e.g. doesn't occur on windows. 
Seems to occur only on linux platforms.  I've found it occurs on both 
lib++ and libstdc++ evironments.  I can't figure out from the test 
results whether the tests are linking from static or shared standard 
libraries.

I didn't realize that this occurs (only on?) configurations which create 
shared libraries which link against static boost.  Mixing shared 
libraries is a known cause of difficulties.  It many cases it seems to 
work, but in many cases it also leads to unfixable failures.

I'm still looking into this.

> 
> Reproduce with:
> git clone git_at_[hidden]:boostorg/boost.git && cd boost && git submodule 
> update --recursive --init
> ./bootstrap.sh && ./b2 variant=release link=static 
> --with-libraries=serialization cxxflags=-fPIC cflags=-fPIC -j2
> mkdir mytest && cd mytest
> Create 3 files:
> 
> |test_multi_singleton.cpp:intf();intg();intmain(intargc,char**){// Make 
> sure symbols are 
> usedif(argc==8)returnf();if(argc==9)returng();}multi_singleton1.cpp:#include<boost/serialization/extended_type_info_typeid.hpp>intf(){return0!=boost::serialization::extended_type_info_typeid<int>::get_const_instance().get_key();}multi_singleton2.cpp:#include<boost/serialization/extended_type_info_typeid.hpp>intg(){// 
> Use different(!) 
> typereturn0!=boost::serialization::extended_type_info_typeid<float>::get_const_instance().get_key();}| 
> 
> 
> export CPATH=$PWD/..
> export LIBRARY_PATH=$PWD/../stage/lib/
> g++ multi_singleton1.cpp -lboost_serialization -fPIC -shared 
> -olibmulti_singleton1.so
> g++ multi_singleton2.cpp -lboost_serialization -fPIC -shared 
> -olibmulti_singleton2.so
> g++ test_multi_singleton.cpp -L. -lmulti_singleton1 -lmulti_singleton2
> LD_LIBRARY_PATH=. ./a.out
> 
> Result: *** Error in `./a.out': double free or corruption (fasttop): 
> 0x0000000001accc20 ***

I'll look into your test.

> 
> This is due to the bug I described and fixed in 
> https://github.com/boostorg/serialization/pull/105.
> An extract from there only showing the bug with `is_destroyed` is 
> https://github.com/boostorg/serialization/pull/110
> 
> In the current state I'm strongly against including the latest master 
> from Boost.Serialization. It would be better, to include the Boost 1.67 
> version of it which "only" contains a memory leak.

We differ on the cause, the fix and the work around.  Objection noted.

Robert Ramey
> 
> Regards, Alex
> 
> _______________________________________________
> Unsubscribe & other changes: 
> http://lists.boost.org/mailman/listinfo.cgi/boost
>