2009/2/12 Ion Gaztañaga
Gaetan Gaumer wrote:
First of all, if you use fixed_managed_shared_memory, you want to
specify the same base address in ever process, otherwise, you're
code won't work.
Is it enough to create the segment with the same name in each process :
eg a call to the line below in each process :
fixed_managed_shared_memory
segment(open_or_create,"sharedPtrSharedMemoryTest", 2048);
You don't have any guarantee that the OS will map it in the same address in
both processes.
Ok, I'll force the mapping to the same address for all the processes.
I understand your solution but I don't know wich method I have to
redefine.
Because the problem is not really the destructor, but rather the
"deleter".
And I don't find how to redefine (and register) the base deleter to call
the derived one.
Could you help me (again...)
Sorry, it was just a fast idea that I haven't elaborated. But I guess this
will be hard to implement. Inheritance and shared memory don't work very
well
I tried to implement that solution by writing a customDeleter inspired
by boost::interprocess::deleter
where the only difference is the operator() :
template void customDeleter::operator()(const pointer &p){
int objectId = detail::get_pointer(p)->getId();
switch (objectId){
case DERIVED_ID:
mp_mngr->destroy_ptr((derived*)(detail::get_pointer(p))); //cast to the
right type
break;
default:
mp_mngr->destroy_ptr(detail::get_pointer(p));
break;
}
}
Then I use it like this :
typedef fixed_managed_shared_memory::segment_manager
segment_manager_t;
typedef allocator
void_allocator;
typedef customDeleter base_Deleter;
typedef shared_ptr base_shared_ptr;
base_shared_ptr baseSP =
base_shared_ptr((dbSharableData*)segment.construct<derived>(anonymous_instance)(DERIVED_ID,3.3)
,segment.get_allocator<void>(),
base_Deleter(base_Deleter::segment_manager_pointer(segment)) );
This seems to work because the right deleter is called, and assert()
in block_header_from_value() don't fail as before.
But my program get locked in a scoped_lock at
boost/interprocess/mem_algo/rbtree_best_fit.hpp:1264 :
template
void rbtree_best_fit::deallocate(void* addr)
{
if(!addr) return;
//-----------------------
boost::interprocess::scoped_lock guard(m_header);
//-----------------------
return this->priv_deallocate(addr);
}
I don't understand how and why there's such a lock as I work currently with
only one process.
Any hint ?
Regards,
Gaëtan
