Multiple Inheritance + Polymorphism Seg Fault

[AzothAmmo]

See discussion on mailing list and related issue #187.

Minimal example provided by David Bond:

#include <cereal/archives/json.hpp>
#include <cereal/types/polymorphic.hpp>
#include <iostream>

struct Base {
    virtual void foo() { } 
    template<class Archive> void serialize(Archive& archive) { } 
};

struct Base2 {
    virtual void foo() { } 
    std::string var;
    template<class Archive> void serialize(Archive& archive) {   
        archive(cereal::make_nvp("var", var));
    }   
};

struct Derived : public Base2, public Base {
    template<class Archive> void serialize(Archive& archive) {   
        archive(cereal::make_nvp("Base2",
                                 cereal::base_class<Base2>(this)),
                cereal::make_nvp("Base",
                                 cereal::base_class<Base>(this)));
    }   
};

CEREAL_REGISTER_TYPE(Derived);

int main() {
    auto ptr = std::make_shared<Derived>();
    cereal::JSONOutputArchive ar(std::cout);
    ar(cereal::make_nvp("ptr", std::dynamic_pointer_cast<Base>(ptr)));

    return 0;
}

[bbouzan]

I have this same bug on my system, is there any workaround?

[AzothAmmo]

No solution yet, I haven't had the time to look into it. Potentially related issue discussed on the mailing list here: https://groups.google.com/forum/#!topic/cerealcpp/ifoi4a36C5k

[Mokon]

@bbouzan on my project where I hit this I simply removed multiple inheritance. (In my case it was simply a output class that I could do that without any trouble.) That isn't really a solution though... But the key here is the internal code is casting a class to a void* then back to another of the types.

[ylobanov]

Hello,
In our project we are about to start implementation of serializatioin, and there is a choice between cereal and boost::serialization. Cereal looks much more attractive then boost, but this bug is an issue for us. I wonder if it is reasonable to wait for the fix or it's better to use boost. Do you have any estimates? Thanks

[AzothAmmo]

All progress is on a mini-hiatus until December, when I'll have time to tackle the various issues. I can't estimate how long fixing this one individual bug will take, but that's the point when I'd start trying to fix it.

If this timeframe won't work for you, take a look at our boost transition page. It may be possible for you to easily switch to cereal once the issue is resolved as we support an identical syntax to that found in boost serialization.

[ylobanov]

Thanks for the information, we'll wait for December

[AzothAmmo]

Small update on this: I've investigated the problem and know what is causing it though I haven't come up with a solution yet. David Bond correctly identified the source of the issue on the mailing list, we're ending up at an incorrect address do a void * cast.

[AzothAmmo]

So we'll basically have to implement something very much like boost's void_cast_register.

[AzothAmmo]

I have this fixed locally, finishing up documentation and cleanup and will push it through soon. This will involve a breaking change for certain polymorphic serialization situations.

[ylobanov]

thanks for updating!