Hi Robert, On Tue, Dec 12, 2006 at 09:20:49AM -0800, Robert Ramey wrote:

So I would try reformlating your example like the following:

...

I tried your suggestions, but on the Suse gcc 3.3.3 I still get the segfault. For now, I will switch platforms/compilers (your library has proven to be very usefull in the past for us) .. regards, Oscar

On Wed, Dec 13, 2006 at 08:29:40AM -0800, Robert Ramey wrote:

I would be very interested in getting to the bottom of the problem. This would likely result in a small tweak which would make the library more portable. Of course I realise that this is not that easy. Note that I test here with gcc 3.3 under cywin. So I would be hopeful that its easily fixable. I don't know what version of boost your using, but it may already be fixed in a later version.

According to the website of Novell, they use 'GCC 3.3.3 enhanced by the SUSE labs team'. I need to a run patch on this system, but I'm not sure whether this patch will affect gcc. I'll try it anyway and run the same tests again. I'll report the results back later (probably tomorow). If you have any suggestions or tests you want me to run on this system just let me know. The boost version is boost 1_33_1. I'll check for a newer version also. Regards, Oscar

Hi Robert, On Thu, Dec 14, 2006 at 09:05:05AM -0800, Robert Ramey wrote:

If you suspect the compiler I would run the whole test suite. It might take overnight - but its much better then waiting for anomolies to appear in your code and trying to track them down on a case by case basis. You can run just the serialization part of the test suite, but you might as well just follow the instructions with 1.33 and just re-run the whole thing. That wil give you table of any and all issues the version of the compiler you're using might have.

I'm still working on running the tests. It looks like the serialization tests all passed, and currently I'm running all tests of the complete source tree. I noticed that a lot of tests in the type_traits have failed, but BOOST_IS_ABSTRACT should solve that, right ? regards, Oscar

Hi Robert, On Wed, Dec 13, 2006 at 08:29:40AM -0800, Robert Ramey wrote:

I would be very interested in getting to the bottom of the problem. This would likely result in a small tweak which would make the library more portable. Of course I realise that this is not that easy. Note that I test here with gcc 3.3 under cywin. So I would be hopeful that its easily fixable. I don't know what version of boost your using, but it may already be fixed in a later version.

It looks like the problem is fixed (or at least, workaround-ed). When I change string serialize(const base * const b) to string serialize(base * const b) the problem goes away (I don't get the segfault anymore). I ran the regression tests on the suse (gcc 3.3.3) machine and the serialization tests all pass (except for test_static_warning which gives a warn, but I guess that is what's supposed to happen). I am going to use the workaround (remove the first 'const') and consider this thread closed. Except when you want to investigate some more, I'll be glad to help, but then I need specific instructions on what to test because this is getting a bit over my head ... For completeness I included the original message once more at the bottom of this post. Regards and thanks for your suggestions, Oscar -- Original Message: currently I am trying to get serialization to work. It worked well when I was using boost 1.32, but since I upgraded to 1.33 I'm having some problems. In short; I have an abstract base class and a derived class. I try to serialize/deserialize the derived class through a pointer to the base +class. This worked well when I was using 1.32. When I upgraded to 1.33 I got the static assert which trapped the 'saving non const error' (I hope this makes sense ..). So I fixed it by following instructions from other posts dealing with the same problem (basically just by changing 'base *b' to +'const base * const b'). The fix resolved the compile error, only now I'm getting a segfault at runtime. The segfault happens during deserialization (when trying to 'reconstruct' the original object). The funny thing is, I've tested it on a system (suse enterprise system 9, gcc 3.3.3) and it chrashes. On another system (gcc version 4.0.3 (Ubuntu 4.0.3-1ubuntu5)) it works fine. I reproduced the problem in a very small program located at the end of this post. Maybe someone can have a look at the code and tell me whether there are obvious problems with my code or not. If I know for sure that my code should run correctly, I'll just move on to an other linux distro .. Thanks in advance, Oscar #include<boost/archive/text_oarchive.hpp> #include<boost/archive/text_iarchive.hpp> #include<boost/serialization/base_object.hpp> #include<boost/serialization/export.hpp> #include<iostream> #include<sstream> using namespace std; class base { public: virtual void do_it() const = 0; private: friend class boost::serialization::access; template<class Archive> void serialize(Archive & ar, const unsigned int){ cout << "serializing base" << endl; } }; class derived: public base { public: derived(int field1, string field2){ _field1 = field1; _field2 = field2; } virtual void do_it() const { cout << "doing it with " << _field1 << " and " << _field2 << endl; } private: friend class boost::serialization::access; derived(){}; virtual ~derived(){}; template<class Archive> void serialize(Archive & ar, const unsigned int){ cout << "derived being serialized" << endl; ar & boost::serialization::base_object<base>(*this); ar & _field1; ar & _field2; } int _field1; string _field2; }; BOOST_IS_ABSTRACT(base); BOOST_CLASS_EXPORT(derived); string serialize(const base * const b){ stringstream str; boost::archive::text_oarchive oa(str); oa << b; return str.str(); } base *deserialize(string s){ stringstream str(s); base *b; boost::archive::text_iarchive ia(str); ia >> b; return b; } int main(){ base *d1 = new derived(52,"a string"); d1->do_it(); string serialized = serialize(d1); base *d2 = deserialize(serialized); d2->do_it(); return 0; }