Create a "holder" object to fix stale reference to array

we were creating a reference to an int array, but then that object no
longer was existing, so was being deleted, but we are still trying to
use it. This uses the python capsule mechanism to stash off the array
and then clean it up

Signed-off-by: Kimball Thurston <kdt3rd@gmail.com>

PyIlmBase/PyImathNumpy/imathnumpymodule.cpp

@@ -44,6 +44,30 @@
 using namespace boost::python;
 using namespace PyImath;
 
+template <typename T>
+struct Holder
+{
+    Holder( T &a ) : m_val( a ) {}
+    static void Cleanup (PyObject *capsule)
+    {
+        Holder* h = static_cast<Holder*> (PyCapsule_GetPointer (capsule, NULL));
+        delete h;
+    }
+    T m_val;
+};
+
+template <typename T>
+static void
+setBaseObject (PyObject* nparr, T& arr)
+{
+    using converter_type = typename reference_existing_object::apply<T*>::type;
+    using holder         = Holder<T>;
+
+    holder* ph = new holder (arr);
+    PyObject* capsule = PyCapsule_New (ph, NULL, holder::Cleanup);
+    PyArray_SetBaseObject ((PyArrayObject*) nparr, capsule);
+}
+
 static
 object 
 arrayToNumpy_float(FloatArray &fa)
@@ -59,8 +83,9 @@ arrayToNumpy_float(FloatArray &fa)
     if (!a) {
         throw_error_already_set();
     }
+    setBaseObject (a, fa);
 
-    object retval = object(handle<>(a));
+    object retval = object (handle<> (a));
     return retval;
 }
 
@@ -81,6 +106,7 @@ arrayToNumpy_V3f(V3fArray &va)
     if (!a) {
         throw_error_already_set();
     }
+    setBaseObject (a, va);
 
     object retval = object(handle<>(a));
     return retval;
@@ -101,6 +127,7 @@ arrayToNumpy_int(IntArray &va)
     if (!a) {
         throw_error_already_set();
     }
+    setBaseObject (a, va);
 
     object retval = object(handle<>(a));
     return retval;