Implement two-phase session insertion

On the first step an uninitialized object is inserted into the cache
as before (this is necessary to handle recursive loading). The second
step is to notify the session that the object has been initialized.
On this second step the session can perform change tracking preparations,
such as make a copy of the object or reset the modification flag.

New test: common/session/custom (implements a custom session that uses
copies to track changes).

1 files changed, 10 insertions, 4 deletions

diff --git a/odb/database.txx b/odb/database.txx
index df58f35..b5f9640 100644
--- a/odb/database.txx
+++ b/odb/database.txx
@@ -36,8 +36,11 @@ namespace odb
 
     object_traits::persist (*this, obj);
 
-    object_traits::reference_cache_traits::insert (
-      *this, reference_cache_type<T>::convert (obj));
+    typename object_traits::reference_cache_traits::position_type p (
+      object_traits::reference_cache_traits::insert (
+        *this, reference_cache_type<T>::convert (obj)));
+
+    object_traits::reference_cache_traits::initialize (p);
 
     return object_traits::id (obj);
   }
@@ -58,8 +61,11 @@ namespace odb
 
     // Get the canonical object pointer and insert it into object cache.
     //
-    object_traits::pointer_cache_traits::insert (
-      *this, pointer_cache_type<pointer_type>::convert (pobj));
+    typename object_traits::pointer_cache_traits::position_type p (
+      object_traits::pointer_cache_traits::insert (
+        *this, pointer_cache_type<pointer_type>::convert (pobj)));
+
+    object_traits::pointer_cache_traits::initialize (p);
 
     return object_traits::id (obj);
   }