Fix an interesting case where we fail to convert a ctuple to a Python…

… tuple on cascaded assignments (test_grammar.py, line 895) because we do not generate a (useless) PyTypeTestNode any more.

Apparently, that type test is still required in order to trigger the Python conversion, which then renders the type test redundant.

Cython/Compiler/ExprNodes.py

@@ -1039,13 +1039,18 @@ def coerce_to(self, dst_type, env):
                 pass
             elif src.constant_result is None:
                 src = NoneNode(src.pos).coerce_to(dst_type, env)
-            elif not src.type.is_pyobject:
-                if dst_type is bytes_type and src.type.is_int:
-                    src = CoerceIntToBytesNode(src, env)
-                else:
-                    src = CoerceToPyTypeNode(src, env, type=dst_type)
-            elif not src.type.subtype_of(dst_type):
-                src = PyTypeTestNode(src, dst_type, env)
+            else:
+                if not src.type.is_pyobject:
+                    if dst_type is bytes_type and src.type.is_int:
+                        src = CoerceIntToBytesNode(src, env)
+                    else:
+                        src = CoerceToPyTypeNode(src, env, type=dst_type)
+                # FIXME: I would expect that CoerceToPyTypeNode(type=dst_type) returns a value of type dst_type
+                #        but it doesn't for ctuples. Thus, we add a PyTypeTestNode which then triggers the
+                #        Python conversion and becomes useless. That sems backwards and inefficient.
+                #        We should not need a PyTypeTestNode after a previous conversion above.
+                if not src.type.subtype_of(dst_type):
+                    src = PyTypeTestNode(src, dst_type, env)
         elif is_pythran_expr(dst_type) and is_pythran_supported_type(src.type):
             # We let the compiler decide whether this is valid
             return src