throw proper error if numpy array that contains object is serialized

python/src/pynumbuf/adapters/numpy.cc

@@ -1,5 +1,7 @@
 #include "numpy.h"
 
+#include <sstream>
+
 #include <numbuf/tensor.h>
 
 using namespace arrow;
@@ -95,7 +97,9 @@ Status SerializeArray(PyArrayObject* array, SequenceBuilder& builder) {
       RETURN_NOT_OK(builder.AppendTensor(dims, reinterpret_cast<double*>(data)));
       break;
     default:
-      DCHECK(false) << "numpy data type not recognized: " << dtype;
+      std::stringstream stream;
+      stream << "numpy data type not recognized: " << dtype;
+      return Status::NotImplemented(stream.str());
   }
   Py_XDECREF(contiguous);
   return Status::OK();

python/test/runtest.py

@@ -47,6 +47,13 @@ class SerializationTests(unittest.TestCase):
     for t in ["int8", "uint8", "int16", "uint16", "int32", "uint32", "float32", "float64"]:
       self.numpyTest(t)
 
+  def testNumpyObject(self):
+    a = np.array([np.zeros((2,2))], dtype=object)
+    try:
+      x = self.roundTripTest([a])
+    except:
+      pass
+
   def testRay(self):
     for obj in TEST_OBJECTS:
       self.roundTripTest([obj])