fix error handling in backported matrix multiplication and compare to actual behaviour in Py3.5

Cython/Utility/ObjectHandling.c

@@ -1353,7 +1353,8 @@ static CYTHON_INLINE PyObject* __Pyx_PyObject_CallNoArg(PyObject *func) {
   #define __Pyx_PyNumber_MatrixMultiply(x,y)         PyNumber_MatrixMultiply(x,y)
   #define __Pyx_PyNumber_InPlaceMatrixMultiply(x,y)  PyNumber_InPlaceMatrixMultiply(x,y)
 #else
-static PyObject* __Pyx_PyNumber_MatrixMultiply(PyObject* x, PyObject* y);
+#define __Pyx_PyNumber_MatrixMultiply(x,y)         __Pyx__PyNumber_MatrixMultiply(x, y, "@")
+static PyObject* __Pyx__PyNumber_MatrixMultiply(PyObject* x, PyObject* y, const char* op_name);
 static PyObject* __Pyx_PyNumber_InPlaceMatrixMultiply(PyObject* x, PyObject* y);
 #endif
 
@@ -1392,7 +1393,7 @@ bad:
     return result;
 }
 
-static PyObject* __Pyx_PyNumber_MatrixMultiply(PyObject* x, PyObject* y) {
+static PyObject* __Pyx__PyNumber_MatrixMultiply(PyObject* x, PyObject* y, const char* op_name) {
     PyObject *func;
     // FIXME: make subtype aware
     // see note at https://docs.python.org/3/reference/datamodel.html#emulating-numeric-types
@@ -1410,10 +1411,20 @@ static PyObject* __Pyx_PyNumber_MatrixMultiply(PyObject* x, PyObject* y) {
     func = __Pyx_PyObject_GetAttrStr(y, PYIDENT("__rmatmul__"));
     if (func) {
         PyObject *result = __Pyx_PyObject_CallMatrixMethod(func, x);
-        return result;
+        if (result != Py_NotImplemented)
+            return result;
+        Py_DECREF(result);
+    } else {
+        if (!PyErr_ExceptionMatches(PyExc_AttributeError))
+            return NULL;
+        PyErr_Clear();
     }
-    Py_INCREF(Py_NotImplemented);
-    return Py_NotImplemented;
+    PyErr_Format(PyExc_TypeError,
+                 "unsupported operand type(s) for %.2s: '%.100s' and '%.100s'",
+                 op_name,
+                 Py_TYPE(x)->tp_name,
+                 Py_TYPE(y)->tp_name);
+    return NULL;
 }
 
 static PyObject* __Pyx_PyNumber_InPlaceMatrixMultiply(PyObject* x, PyObject* y) {
@@ -1429,6 +1440,6 @@ static PyObject* __Pyx_PyNumber_InPlaceMatrixMultiply(PyObject* x, PyObject* y)
             return NULL;
         PyErr_Clear();
     }
-    return __Pyx_PyNumber_MatrixMultiply(x, y);
+    return __Pyx__PyNumber_MatrixMultiply(x, y, "@=");
 }
 #endif

tests/run/matrix_multiplier.pyx

@@ -17,6 +17,28 @@ ExtMatMult(1) @ 22
 ExtMatMult('ExtMatMult(1) @ ExtMatMult(2)')
 >>> print(test_imatmul(a, b))
 ExtMatMult("ExtMatMult('ExtMatMult(1) @ ExtMatMult(2)') @ ExtMatMult(2)")
+
+>>> x = y = 1
+>>> x @ y
+Traceback (most recent call last):
+TypeError: unsupported operand type(s) for @: 'int' and 'int'
+>>> x @= y
+Traceback (most recent call last):
+TypeError: unsupported operand type(s) for @=: 'int' and 'int'
+
+>>> y = MatMult(22)
+>>> x @= y
+>>> print(x)
+1 @ MatMult(22)
+
+>>> x = MatMult(22)
+>>> print(x @ 1)
+MatMult(22) @ 1
+>>> print(1 @ x)
+1 @ MatMult(22)
+>>> x @= 1
+>>> print(x)
+MatMult('MatMult(22) @ 1')
 """
 
 
@@ -71,6 +93,10 @@ def test_matmul(a, b):
     11 @ MatMult(2)
     >>> print(test_matmul(MatMult('abc'), MatMult('def')))
     MatMult('abc') @ MatMult('def')
+
+    >>> test_matmul(1, 2)
+    Traceback (most recent call last):
+    TypeError: unsupported operand type(s) for @: 'int' and 'int'
     """
     return a @ b
 
@@ -81,6 +107,14 @@ def test_imatmul(a, b):
     MatMult('MatMult(1) @ MatMult(2)')
     >>> print(test_imatmul(MatMult('abc'), MatMult('def')))
     MatMult("MatMult('abc') @ MatMult('def')")
+    >>> print(test_imatmul(11, MatMult('def')))
+    11 @ MatMult('def')
+    >>> print(test_imatmul(MatMult('abc'), 11))
+    MatMult("MatMult('abc') @ 11")
+
+    >>> test_imatmul(1, 2)
+    Traceback (most recent call last):
+    TypeError: unsupported operand type(s) for @=: 'int' and 'int'
     """
     a @= b
     return a