[Boost-users] boost::interprocess::node_allocator & VC8 run-time problem

I have the problem with VC8 release build (all release build options - by default, i.e., optimization is "Maximize speed /O2" and so on). A run of "repro" output executable finishes with the exception on attempt to read data from address 0. I noticed it for boost::interprocess (and boost::shmem too) allocators of type: - node_allocator - cached_node_allocator and not to: - allocator - private_node_allocator The way out I found (for "problematic" allocator types) is to set additional compiler option "/Og-" (already declared deprecated). With VC7.1 it all is ok for any allocator type w/o setting /Og- or any other additional options. Have anyone noticed this problem? Where the problem originates from? VC8 optimization specifics (NRVO is affected by /Og option)? Or smth. wrong in "repro" code? Thanks. ////////////// The "repro" code is the following: #include "stdafx.h" #include #include #include using namespace boost::interprocess; enum { A_SHARED_MEMORY_SIZE = 65536, A_POOL_SIZE = 64 }; typedef node_allocator user_allocator_t; typedef list MyUserList; int _tmain(int argc, _TCHAR* argv[]) { static char static_buffer[A_SHARED_MEMORY_SIZE]; managed_external_buffer user_buffer(create_only, static_buffer, A_SHARED_MEMORY_SIZE); MyUserList *userlist = user_buffer.construct<MyUserList>("MyUserList")(user_buffer.get_segment_manager()); return 0; }