Fixing constructor bug pytensor<..., 0>

Author: tdegeus

include/xtensor-python/pytensor.hpp

@@ -100,11 +100,26 @@ namespace pybind11
             }
         };
 
-    }
+    } // namespace detail
 }
 
 namespace xt
 {
+    namespace detail {
+
+        template <std::size_t N, typename = void>
+        struct numpy_strides
+        {
+            npy_intp value[N];
+        };
+
+        template <std::size_t N>
+        struct numpy_strides<N, typename std::enable_if_t<(N == 0)>::type>
+        {
+            npy_intp* value = nullptr;
+        };
+
+    } // namespace detail
 
     template <class T, std::size_t N, layout_type L>
     struct xiterable_inner_types<pytensor<T, N, L>>
@@ -433,8 +448,8 @@ namespace xt
     template <class T, std::size_t N, layout_type L>
     inline void pytensor<T, N, L>::init_tensor(const shape_type& shape, const strides_type& strides)
     {
-        npy_intp python_strides[N];
-        std::transform(strides.begin(), strides.end(), python_strides,
+        detail::numpy_strides<N> python_strides;
+        std::transform(strides.begin(), strides.end(), python_strides.value,
                        [](auto v) { return sizeof(T) * v; });
         int flags = NPY_ARRAY_ALIGNED;
         if (!std::is_const<T>::value)
@@ -445,7 +460,7 @@ namespace xt
 
         auto tmp = pybind11::reinterpret_steal<pybind11::object>(
             PyArray_NewFromDescr(&PyArray_Type, (PyArray_Descr*) dtype.release().ptr(), static_cast<int>(shape.size()),
-                        const_cast<npy_intp*>(shape.data()), python_strides,
+                        const_cast<npy_intp*>(shape.data()), python_strides.value,
                         nullptr, flags, nullptr));
 
         if (!tmp)

test/test_pytensor.cpp

@@ -65,6 +65,15 @@ namespace xt
         EXPECT_THROW(pyt3::from_shape(shp), std::runtime_error);
     }
 
+    TEST(pytensor, scalar_from_shape)
+    {
+        std::array<size_t, 0> shape;
+        auto a = pytensor<double, 0>::from_shape(shape);
+        pytensor<double, 0> b(1.2);
+        EXPECT_TRUE(a.size() == b.size());
+        EXPECT_TRUE(xt::has_shape(a, b.shape()));
+    }
+
     TEST(pytensor, strided_constructor)
     {
         central_major_result<container_type> cmr;

test_python/main.cpp

@@ -227,6 +227,11 @@ void col_major_array(xt::pyarray<double, xt::layout_type::column_major>& arg)
     }
 }
 
+xt::pytensor<int, 0> xscalar(const xt::pytensor<int, 1>& arg)
+{
+    return xt::sum(arg);
+}
+
 template <class T>
 using ndarray = xt::pyarray<T, xt::layout_type::row_major>;
 
@@ -285,6 +290,8 @@ PYBIND11_MODULE(xtensor_python_test, m)
     m.def("col_major_array", col_major_array);
     m.def("row_major_tensor", row_major_tensor);
 
+    m.def("xscalar", xscalar);
+
     py::class_<C>(m, "C")
         .def(py::init<>())
         .def_property_readonly(

test_python/test_pyarray.py

@@ -151,6 +151,10 @@ def test_col_row_major(self):
         xt.col_major_array(varF)
         xt.col_major_array(varF[:, :, 0]) # still col major!
 
+    def test_xscalar(self):
+        var = np.arange(50, dtype=int)
+        self.assertTrue(np.sum(var) == xt.xscalar(var))
+
     def test_bad_argument_call(self):
         with self.assertRaises(TypeError):
             xt.simple_array("foo")