diff --git a/include/boost/mpi/collectives/all_gather.hpp b/include/boost/mpi/collectives/all_gather.hpp
index d44ea42..da665e9 100644
--- a/include/boost/mpi/collectives/all_gather.hpp
+++ b/include/boost/mpi/collectives/all_gather.hpp
@@ -27,7 +27,7 @@ namespace detail {
   all_gather_impl(const communicator& comm, const T* in_values, int n, 
                   T* out_values, mpl::true_)
   {
-    MPI_Datatype type = boost::mpi::get_mpi_datatype<T>();
+    MPI_Datatype type = boost::mpi::get_mpi_datatype<T>(*in_values);
     BOOST_MPI_CHECK_RESULT(MPI_Allgather,
                            (const_cast<T*>(in_values), n, type,
                             out_values, n, type, comm));
diff --git a/include/boost/mpi/collectives/all_reduce.hpp b/include/boost/mpi/collectives/all_reduce.hpp
index 41acdb9..fbb4090 100644
--- a/include/boost/mpi/collectives/all_reduce.hpp
+++ b/include/boost/mpi/collectives/all_reduce.hpp
@@ -32,7 +32,7 @@ namespace detail {
   {
     BOOST_MPI_CHECK_RESULT(MPI_Allreduce,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             is_mpi_op<Op, T>::op(), comm));
   }
 
@@ -51,7 +51,7 @@ namespace detail {
     user_op<Op, T> mpi_op(op);
     BOOST_MPI_CHECK_RESULT(MPI_Allreduce,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             mpi_op.get_mpi_op(), comm));
   }
 
diff --git a/include/boost/mpi/collectives/all_to_all.hpp b/include/boost/mpi/collectives/all_to_all.hpp
index 61f512b..8c33c2a 100644
--- a/include/boost/mpi/collectives/all_to_all.hpp
+++ b/include/boost/mpi/collectives/all_to_all.hpp
@@ -30,7 +30,7 @@ namespace detail {
   all_to_all_impl(const communicator& comm, const T* in_values, int n, 
                   T* out_values, mpl::true_)
   {
-    MPI_Datatype type = get_mpi_datatype<T>();
+    MPI_Datatype type = get_mpi_datatype<T>(*in_values);
     BOOST_MPI_CHECK_RESULT(MPI_Alltoall,
                            (const_cast<T*>(in_values), n, type,
                             out_values, n, type, comm));
@@ -87,6 +87,10 @@ namespace detail {
     }
     std::vector<char, allocator<char> > incoming(sum > 0? sum : 1);
 
+    // Make sure we don't try to reference an empty vector
+    if (outgoing.empty())
+      outgoing.push_back(0);
+
     // Transmit the actual data
     BOOST_MPI_CHECK_RESULT(MPI_Alltoallv,
                            (&outgoing[0], &send_sizes[0],
diff --git a/include/boost/mpi/collectives/broadcast.hpp b/include/boost/mpi/collectives/broadcast.hpp
index 7a8ce83..a615cc2 100644
--- a/include/boost/mpi/collectives/broadcast.hpp
+++ b/include/boost/mpi/collectives/broadcast.hpp
@@ -95,7 +95,7 @@ namespace detail {
   {
     BOOST_MPI_CHECK_RESULT(MPI_Bcast,
                            (values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*values),
                             root, MPI_Comm(comm)));
   }
 
diff --git a/include/boost/mpi/collectives/gather.hpp b/include/boost/mpi/collectives/gather.hpp
index ab104bd..5c605a8 100644
--- a/include/boost/mpi/collectives/gather.hpp
+++ b/include/boost/mpi/collectives/gather.hpp
@@ -28,7 +28,7 @@ namespace detail {
   gather_impl(const communicator& comm, const T* in_values, int n, 
               T* out_values, int root, mpl::true_)
   {
-    MPI_Datatype type = get_mpi_datatype<T>();
+    MPI_Datatype type = get_mpi_datatype<T>(*in_values);
     BOOST_MPI_CHECK_RESULT(MPI_Gather,
                            (const_cast<T*>(in_values), n, type,
                             out_values, n, type, root, comm));
@@ -41,7 +41,7 @@ namespace detail {
   gather_impl(const communicator& comm, const T* in_values, int n, int root, 
               mpl::true_)
   {
-    MPI_Datatype type = get_mpi_datatype<T>();
+    MPI_Datatype type = get_mpi_datatype<T>(*in_values);
     BOOST_MPI_CHECK_RESULT(MPI_Gather,
                            (const_cast<T*>(in_values), n, type,
                             0, n, type, root, comm));
@@ -99,10 +99,12 @@ void
 gather(const communicator& comm, const T& in_value, std::vector<T>& out_values,
        int root)
 {
-  if (comm.rank() == root)
+  if (comm.rank() == root) {
     out_values.resize(comm.size());
-
-  ::boost::mpi::gather(comm, in_value, &out_values[0], root);
+    ::boost::mpi::gather(comm, in_value, &out_values[0], root);
+  } else {
+    ::boost::mpi::gather(comm, in_value, root);
+  }
 }
 
 template<typename T>
diff --git a/include/boost/mpi/collectives/reduce.hpp b/include/boost/mpi/collectives/reduce.hpp
index 39c18de..6ea8279 100644
--- a/include/boost/mpi/collectives/reduce.hpp
+++ b/include/boost/mpi/collectives/reduce.hpp
@@ -51,7 +51,7 @@ namespace detail {
   {
     BOOST_MPI_CHECK_RESULT(MPI_Reduce,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             is_mpi_op<Op, T>::op(), root, comm));
   }
 
@@ -64,7 +64,7 @@ namespace detail {
   {
     BOOST_MPI_CHECK_RESULT(MPI_Reduce,
                            (const_cast<T*>(in_values), 0, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             is_mpi_op<Op, T>::op(), root, comm));
   }
 
@@ -84,7 +84,7 @@ namespace detail {
     user_op<Op, T> mpi_op(op);
     BOOST_MPI_CHECK_RESULT(MPI_Reduce,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             mpi_op.get_mpi_op(), root, comm));
   }
 
@@ -99,7 +99,7 @@ namespace detail {
     user_op<Op, T> mpi_op(op);
     BOOST_MPI_CHECK_RESULT(MPI_Reduce,
                            (const_cast<T*>(in_values), 0, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             mpi_op.get_mpi_op(), root, comm));
   }
 
diff --git a/include/boost/mpi/collectives/scan.hpp b/include/boost/mpi/collectives/scan.hpp
index e01359a..83dc8ca 100644
--- a/include/boost/mpi/collectives/scan.hpp
+++ b/include/boost/mpi/collectives/scan.hpp
@@ -51,7 +51,7 @@ namespace detail {
   {
     BOOST_MPI_CHECK_RESULT(MPI_Scan,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             is_mpi_op<Op, T>::op(), comm));
   }
 
@@ -70,7 +70,7 @@ namespace detail {
     user_op<Op, T> mpi_op(op);
     BOOST_MPI_CHECK_RESULT(MPI_Scan,
                            (const_cast<T*>(in_values), out_values, n,
-                            boost::mpi::get_mpi_datatype<T>(),
+                            boost::mpi::get_mpi_datatype<T>(*in_values),
                             mpi_op.get_mpi_op(), comm));
   }
 
diff --git a/include/boost/mpi/collectives/scatter.hpp b/include/boost/mpi/collectives/scatter.hpp
index 125d001..4d9ed8e 100644
--- a/include/boost/mpi/collectives/scatter.hpp
+++ b/include/boost/mpi/collectives/scatter.hpp
@@ -28,7 +28,7 @@ namespace detail {
   scatter_impl(const communicator& comm, const T* in_values, T* out_values, 
                int n, int root, mpl::true_)
   {
-    MPI_Datatype type = get_mpi_datatype<T>();
+    MPI_Datatype type = get_mpi_datatype<T>(*in_values);
     BOOST_MPI_CHECK_RESULT(MPI_Scatter,
                            (const_cast<T*>(in_values), n, type,
                             out_values, n, type, root, comm));
@@ -41,7 +41,7 @@ namespace detail {
   scatter_impl(const communicator& comm, T* out_values, int n, int root, 
                mpl::true_)
   {
-    MPI_Datatype type = get_mpi_datatype<T>();
+    MPI_Datatype type = get_mpi_datatype<T>(*out_values);
     BOOST_MPI_CHECK_RESULT(MPI_Scatter,
                            (0, n, type,
                             out_values, n, type,
@@ -112,10 +112,10 @@ scatter(const communicator& comm, const std::vector<T>& in_values, T& out_value,
         int root)
 {
   if (comm.rank() == root)
-    ::boost::mpi::scatter(comm, &in_values[0], out_value, root);
+    ::boost::mpi::scatter<T>(comm, &in_values[0], out_value, root);
   else
-    ::boost::mpi::scatter(comm, static_cast<const T*>(0), out_value, 
-                                    root);
+    ::boost::mpi::scatter<T>(comm, static_cast<const T*>(0), out_value, 
+                             root);
 }
 
 template<typename T>
diff --git a/include/boost/mpi/communicator.hpp b/include/boost/mpi/communicator.hpp
index 9362ec9..d0bf763 100644
--- a/include/boost/mpi/communicator.hpp
+++ b/include/boost/mpi/communicator.hpp
@@ -35,6 +35,11 @@
 #include <boost/mpi/status.hpp>
 #include <boost/mpi/request.hpp>
 
+#ifdef BOOST_MSVC
+#  pragma warning(push)
+#  pragma warning(disable : 4800) // forcing to bool 'true' or 'false'
+#endif
+
 namespace boost { namespace mpi {
 
 /**
@@ -955,7 +960,7 @@ optional<int> status::count_impl(mpl::true_)
 
   int return_value;
   BOOST_MPI_CHECK_RESULT(MPI_Get_count,
-                         (&m_status, get_mpi_datatype<T>(), &return_value));
+                         (&m_status, get_mpi_datatype<T>(T()), &return_value));
   if (return_value == MPI_UNDEFINED)
     return optional<int>();
   else
@@ -979,7 +984,7 @@ void
 communicator::send_impl(int dest, int tag, const T& value, mpl::true_) const
 {
   BOOST_MPI_CHECK_RESULT(MPI_Send,
-                         (const_cast<T*>(&value), 1, get_mpi_datatype<T>(),
+                         (const_cast<T*>(&value), 1, get_mpi_datatype<T>(value),
                           dest, tag, MPI_Comm(*this)));
 }
 
@@ -1011,7 +1016,8 @@ communicator::array_send_impl(int dest, int tag, const T* values, int n,
                               mpl::true_) const
 {
   BOOST_MPI_CHECK_RESULT(MPI_Send,
-                         (const_cast<T*>(values), n, get_mpi_datatype<T>(),
+                         (const_cast<T*>(values), n, 
+						  get_mpi_datatype<T>(*values),
                           dest, tag, MPI_Comm(*this)));
 }
 
@@ -1043,7 +1049,8 @@ status communicator::recv_impl(int source, int tag, T& value, mpl::true_) const
   status stat;
 
   BOOST_MPI_CHECK_RESULT(MPI_Recv,
-                         (const_cast<T*>(&value), 1, get_mpi_datatype<T>(),
+                         (const_cast<T*>(&value), 1, 
+						  get_mpi_datatype<T>(value),
                           source, tag, MPI_Comm(*this), &stat.m_status));
   return stat;
 }
@@ -1077,7 +1084,8 @@ communicator::array_recv_impl(int source, int tag, T* values, int n,
 {
   status stat;
   BOOST_MPI_CHECK_RESULT(MPI_Recv,
-                         (const_cast<T*>(values), n, get_mpi_datatype<T>(),
+                         (const_cast<T*>(values), n, 
+						  get_mpi_datatype<T>(*values),
                           source, tag, MPI_Comm(*this), &stat.m_status));
   return stat;
 }
@@ -1123,7 +1131,8 @@ communicator::isend_impl(int dest, int tag, const T& value, mpl::true_) const
 {
   request req;
   BOOST_MPI_CHECK_RESULT(MPI_Isend,
-                         (const_cast<T*>(&value), 1, get_mpi_datatype<T>(),
+                         (const_cast<T*>(&value), 1, 
+						  get_mpi_datatype<T>(value),
                           dest, tag, MPI_Comm(*this), &req.m_requests[0]));
   return req;
 }
@@ -1157,7 +1166,8 @@ communicator::array_isend_impl(int dest, int tag, const T* values, int n,
 {
   request req;
   BOOST_MPI_CHECK_RESULT(MPI_Isend,
-                         (const_cast<T*>(values), n, get_mpi_datatype<T>(),
+                         (const_cast<T*>(values), n, 
+						  get_mpi_datatype<T>(*values),
                           dest, tag, MPI_Comm(*this), &req.m_requests[0]));
   return req;
 }
@@ -1396,7 +1406,8 @@ communicator::irecv_impl(int source, int tag, T& value, mpl::true_) const
 {
   request req;
   BOOST_MPI_CHECK_RESULT(MPI_Irecv,
-                         (const_cast<T*>(&value), 1, get_mpi_datatype<T>(),
+                         (const_cast<T*>(&value), 1, 
+						  get_mpi_datatype<T>(value),
                           source, tag, MPI_Comm(*this), &req.m_requests[0]));
   return req;
 }
@@ -1412,7 +1423,8 @@ communicator::irecv_impl(int source, int tag, T& value, mpl::false_) const
   req.m_handler = request::handle_serialized_irecv<T>;
 
   BOOST_MPI_CHECK_RESULT(MPI_Irecv,
-                         (&data->count, 1, get_mpi_datatype<std::size_t>(),
+                         (&data->count, 1, 
+						  get_mpi_datatype<std::size_t>(data->count),
                           source, tag, MPI_Comm(*this), &req.m_requests[0]));
   
   return req;
@@ -1432,7 +1444,8 @@ communicator::array_irecv_impl(int source, int tag, T* values, int n,
 {
   request req;
   BOOST_MPI_CHECK_RESULT(MPI_Irecv,
-                         (const_cast<T*>(values), n, get_mpi_datatype<T>(),
+                         (const_cast<T*>(values), n, 
+						  get_mpi_datatype<T>(*values),
                           source, tag, MPI_Comm(*this), &req.m_requests[0]));
   return req;
 }
@@ -1449,7 +1462,8 @@ communicator::array_irecv_impl(int source, int tag, T* values, int n,
   req.m_handler = request::handle_serialized_array_irecv<T>;
 
   BOOST_MPI_CHECK_RESULT(MPI_Irecv,
-                         (&data->count, 1, get_mpi_datatype<std::size_t>(),
+                         (&data->count, 1, 
+						  get_mpi_datatype<std::size_t>(data->count),
                           source, tag, MPI_Comm(*this), &req.m_requests[0]));
 
   return req;
@@ -1580,4 +1594,8 @@ communicator::irecv<content>(int source, int tag,
 #  include <boost/mpi/detail/communicator_sc.hpp>
 #endif
 
+#ifdef BOOST_MSVC
+#  pragma warning(pop)
+#endif
+
 #endif // BOOST_MPI_COMMUNICATOR_HPP
diff --git a/include/boost/mpi/config.hpp b/include/boost/mpi/config.hpp
index 5bcc271..92c2c83 100644
--- a/include/boost/mpi/config.hpp
+++ b/include/boost/mpi/config.hpp
@@ -39,6 +39,22 @@
 #  define BOOST_MPI_HAS_NOARG_INITIALIZATION
 #endif
 
+#if defined(MPIAPI)
+#  define BOOST_MPI_CALLING_CONVENTION MPIAPI
+#else
+/** @brief Specifies the calling convention that will be used for callbacks
+ *  from the underlying C MPI. 
+ *
+ *  This is a Windows-specific macro, which will be used internally to state
+ *  the calling convention of any function that is to be used as a callback
+ *  from MPI. For example, the internally-defined functions that are used in
+ *  a call to @c MPI_Op_create. This macro is likely only to be useful to
+ *  users that wish to bypass Boost.MPI, registering their own callbacks in
+ *  certain cases, e.g., through @c MPI_Op_create.
+ */
+#  define BOOST_MPI_CALLING_CONVENTION
+#endif
+
 #if defined(LAM_MPI)
 // Configuration for LAM/MPI
 #  define BOOST_MPI_HAS_MEMORY_ALLOCATION
diff --git a/include/boost/mpi/detail/mpi_datatype_cache.hpp b/include/boost/mpi/detail/mpi_datatype_cache.hpp
index dc98911..4f19970 100644
--- a/include/boost/mpi/detail/mpi_datatype_cache.hpp
+++ b/include/boost/mpi/detail/mpi_datatype_cache.hpp
@@ -18,6 +18,15 @@
 #include <map>
 #include <typeinfo>
 
+// The std::type_info::before function in Visual C++ 8.0 (and probably earlier)
+// incorrectly returns an "int" instead of a "bool". Then the compiler has the
+// audacity to complain when that "int" is converted to a "bool". Silence
+// this warning.
+#ifdef BOOST_MSVC
+#  pragma warning(push)
+#  pragma warning(disable : 4800)
+#endif
+
 namespace boost { namespace mpi { namespace detail {
 
 /// @brief comparison function object for two std::type_info pointers
@@ -55,13 +64,13 @@ public:
   }
 
   template <class T>
-  MPI_Datatype datatype(const T& = T(), typename boost::enable_if<is_mpi_builtin_datatype<T> >::type* =0)
+  MPI_Datatype datatype(const T& x = T(), typename boost::enable_if<is_mpi_builtin_datatype<T> >::type* =0)
   {
-    return get_mpi_datatype<T>();
+    return get_mpi_datatype<T>(x);
   }
 
   template <class T>
-  MPI_Datatype datatype(const T& x=T(), typename boost::disable_if<is_mpi_builtin_datatype<T> >::type* =0 )
+  MPI_Datatype datatype(const T& x =T(), typename boost::disable_if<is_mpi_builtin_datatype<T> >::type* =0 )
   {
     BOOST_MPL_ASSERT((is_mpi_datatype<T>));
 
@@ -94,5 +103,8 @@ extern mpi_datatype_map mpi_datatype_cache;
 
 } } } // end namespace boost::mpi::detail
 
+#ifdef BOOST_MSVC
+#  pragma warning(pop)
+#endif
 
 #endif // BOOST_MPI_DETAIL_TYPE_MPI_DATATYPE_CACHE_HPP
diff --git a/include/boost/mpi/detail/packed_oprimitive.hpp b/include/boost/mpi/detail/packed_oprimitive.hpp
index 9cae7fa..b752cf0 100644
--- a/include/boost/mpi/detail/packed_oprimitive.hpp
+++ b/include/boost/mpi/detail/packed_oprimitive.hpp
@@ -74,7 +74,7 @@ public:
     template<class T>
     void save(const T & t)
     {
-        save_impl(&t, get_mpi_datatype<T>(), 1);
+        save_impl(&t, get_mpi_datatype<T>(t), 1);
     }
 
         void save(const std::string &s)
diff --git a/include/boost/mpi/operations.hpp b/include/boost/mpi/operations.hpp
index edb326c..c1189e4 100644
--- a/include/boost/mpi/operations.hpp
+++ b/include/boost/mpi/operations.hpp
@@ -18,7 +18,7 @@
 #ifndef BOOST_MPI_IS_MPI_OP_HPP
 #define BOOST_MPI_IS_MPI_OP_HPP
 
-#include <mpi.h>
+#include <boost/mpi/config.hpp>
 #include <boost/mpl/bool.hpp>
 #include <boost/mpl/if.hpp>
 #include <boost/mpl/and.hpp>
@@ -305,7 +305,7 @@ namespace detail {
     MPI_Op mpi_op;
     static Op* op_ptr;
 
-    static void perform(void* vinvec, void* voutvec, int* plen, MPI_Datatype*)
+    static void BOOST_MPI_CALLING_CONVENTION perform(void* vinvec, void* voutvec, int* plen, MPI_Datatype*)
     {
       T* invec = static_cast<T*>(vinvec);
       T* outvec = static_cast<T*>(voutvec);