I just came across the following bug http://boost.2283326.n4.nabble.com/Serialization-Bizarre-bug-td2612558.html and noticed that the bizarre behaviour is still reproducible (I used boost-version 1.52). I could figure out that in the buggy build there is an additional static initialization of the singlenton<T>::instance in boost\serialization\singleton.hpp in line 149 at startup

serial_bug.exe!`dynamic initializer for 'boost::serialization::singleton<boost::archive::detail::pointer_oserializer<boost::archive::text_oarchive,std::vector<char,std::allocator<char>

...

...

::instance''() Zeile 149 C++

which leads to a call of set_bpis in iserializer.hpp line 347. Thus, std::vector<char> will be saved (but NOT loaded) through a pointer. For apps that store and load the same types the problem usually does not arise, but for server/clients this may not be the case (e.g. apps using the RCF-library) Tobias -- View this message in context: http://boost.2283326.n4.nabble.com/Serialization-Bizarre-bug-reviewed-tp4641... Sent from the Boost - Dev mailing list archive at Nabble.com.