fix bugs of https://github.com/neopenx/Dragon/issues/6

Dragon/modules/python/dragon.cc

@@ -58,7 +58,7 @@ PyObject* RegisteredOperatorsCC(PyObject* self, PyObject* args) {
     PyObject* list = PyList_New(all_keys.size());
     PyObject* list = PyList_New(all_keys.size());
     int idx = 0;
     int idx = 0;
     for (const string& name : all_keys) 
     for (const string& name : all_keys) 
-        CHECK_EQ(PyList_SetItem(list, idx++, StdStringToPyBytes(name)), 0);
+        CHECK_EQ(PyList_SetItem(list, idx++, StdStringToPyUnicode(name)), 0);
     return list;
     return list;
 }
 }
 
 
@@ -68,7 +68,7 @@ PyObject* NoGradientOperatorsCC(PyObject* self, PyObject* args) {
     PyObject* list = PyList_New(all_keys.size());
     PyObject* list = PyList_New(all_keys.size());
     int idx = 0;
     int idx = 0;
     for (const string& name : all_keys) 
     for (const string& name : all_keys) 
-        CHECK_EQ(PyList_SetItem(list, idx++, StdStringToPyBytes(name)), 0);
+        CHECK_EQ(PyList_SetItem(list, idx++, StdStringToPyUnicode(name)), 0);
     return list;
     return list;
 }
 }
 
 
@@ -106,7 +106,7 @@ PyObject* CreateGradientDefsCC(PyObject* self, PyObject* args) {
 
 
     PyObject* g_input_py = PyList_New(grad.g_inputs.size());
     PyObject* g_input_py = PyList_New(grad.g_inputs.size());
     for (int i = 0; i < grad.g_inputs.size(); i++) 
     for (int i = 0; i < grad.g_inputs.size(); i++) 
-        CHECK_EQ(PyList_SetItem(g_input_py, i, StdStringToPyBytes(grad.g_inputs[i])), 0);
+        CHECK_EQ(PyList_SetItem(g_input_py, i, StdStringToPyUnicode(grad.g_inputs[i])), 0);
 
 
     PyObject* defaults_py = PyList_New(grad.defaults.size());
     PyObject* defaults_py = PyList_New(grad.defaults.size());
     for (int i = 0; i < grad.defaults.size(); i++) 
     for (int i = 0; i < grad.defaults.size(); i++) 
@@ -149,14 +149,14 @@ PyObject* SwitchWorkspaceCC(PyObject* self, PyObject *args) {
 }
 }
 
 
 PyObject* CurrentWorkspaceCC(PyObject* self, PyObject* args) { 
 PyObject* CurrentWorkspaceCC(PyObject* self, PyObject* args) { 
-    return StdStringToPyBytes(g_current_workspace);
+    return StdStringToPyUnicode(g_current_workspace);
 }
 }
 
 
 PyObject* WorkspacesCC(PyObject* self, PyObject* args) {
 PyObject* WorkspacesCC(PyObject* self, PyObject* args) {
     PyObject* list = PyList_New(g_workspaces.size());
     PyObject* list = PyList_New(g_workspaces.size());
     int i = 0;
     int i = 0;
     for (auto const& it : g_workspaces) 
     for (auto const& it : g_workspaces) 
-        CHECK_EQ(PyList_SetItem(list, i++, StdStringToPyBytes(it.first)), 0);
+        CHECK_EQ(PyList_SetItem(list, i++, StdStringToPyUnicode(it.first)), 0);
     return list;
     return list;
 }
 }
 
 
@@ -176,14 +176,14 @@ PyObject* ResetWorkspaceCC(PyObject* self, PyObject* args) {
 }
 }
 
 
 PyObject* RootFolderCC(PyObject* self, PyObject* args) {
 PyObject* RootFolderCC(PyObject* self, PyObject* args) {
-    return StdStringToPyBytes(g_workspace->GetRootFolder());
+    return StdStringToPyUnicode(g_workspace->GetRootFolder());
 }
 }
 
 
 PyObject* TensorsCC(PyObject* self, PyObject* args) {
 PyObject* TensorsCC(PyObject* self, PyObject* args) {
     vector<string> tensor_strings = g_workspace->GetTensors();
     vector<string> tensor_strings = g_workspace->GetTensors();
     PyObject* list = PyList_New(tensor_strings.size());
     PyObject* list = PyList_New(tensor_strings.size());
     for (int i = 0; i < tensor_strings.size(); i++)
     for (int i = 0; i < tensor_strings.size(); i++)
-        CHECK_EQ(PyList_SetItem(list, i, StdStringToPyBytes(tensor_strings[i])), 0);
+        CHECK_EQ(PyList_SetItem(list, i, StdStringToPyUnicode(tensor_strings[i])), 0);
     return list;
     return list;
 }
 }
 
 
@@ -224,7 +224,7 @@ PyObject* GetTensorNameCC(PyObject* self, PyObject* args) {
     char* cname;
     char* cname;
     if (!PyArg_ParseTuple(args, "s", &cname)) return nullptr;
     if (!PyArg_ParseTuple(args, "s", &cname)) return nullptr;
     string query = g_workspace->GetTensorName(string(cname));
     string query = g_workspace->GetTensorName(string(cname));
-    return StdStringToPyBytes(query);
+    return StdStringToPyUnicode(query);
 }
 }
 
 
 PyObject* CreateGraphCC(PyObject* self, PyObject* args) {
 PyObject* CreateGraphCC(PyObject* self, PyObject* args) {
@@ -263,7 +263,7 @@ PyObject* GraphsCC(PyObject* self, PyObject* args) {
     vector<string> graph_string = g_workspace->GetGraphs();
     vector<string> graph_string = g_workspace->GetGraphs();
     PyObject* list = PyList_New(graph_string.size());
     PyObject* list = PyList_New(graph_string.size());
     for (int i = 0; i < graph_string.size(); i++)
     for (int i = 0; i < graph_string.size(); i++)
-        CHECK_EQ(PyList_SetItem(list, i, StdStringToPyBytes(graph_string[i])), 0);
+        CHECK_EQ(PyList_SetItem(list, i, StdStringToPyUnicode(graph_string[i])), 0);
     return list;
     return list;
 }
 }
 
 

Dragon/modules/python/dragon.h

@@ -21,7 +21,6 @@
 
 
 #ifdef WITH_PYTHON3
 #ifdef WITH_PYTHON3
 #define PyString_AsString PyUnicode_AsUTF8
 #define PyString_AsString PyUnicode_AsUTF8
-#define PyBytes_FromStringAndSize PyUnicode_FromStringAndSize
 #endif
 #endif
 
 
 using namespace dragon;
 using namespace dragon;
@@ -33,6 +32,15 @@ inline std::string PyBytesToStdString(PyObject* pystring) {
 inline PyObject* StdStringToPyBytes(const std::string& str) {
 inline PyObject* StdStringToPyBytes(const std::string& str) {
     return PyBytes_FromStringAndSize(str.c_str(), str.size());
     return PyBytes_FromStringAndSize(str.c_str(), str.size());
 }
 }
+
+inline PyObject* StdStringToPyUnicode(const std::string& str) {
+#ifdef WITH_PYTHON3
+    return PyUnicode_FromStringAndSize(str.c_str(), str.size());
+#else
+    return PyBytes_FromStringAndSize(str.c_str(), str.size());
+#endif
+}
+
 template <typename T>
 template <typename T>
 inline void MakeStringInternal(std::stringstream& ss, const T& t) { ss << t; }
 inline void MakeStringInternal(std::stringstream& ss, const T& t) { ss << t; }
 
 
@@ -114,7 +122,7 @@ class StringFetcher : public TensorFetcherBase {
  public:
  public:
     PyObject* Fetch(const Tensor& tensor) override {
     PyObject* Fetch(const Tensor& tensor) override {
         CHECK_GT(tensor.count(), 0);
         CHECK_GT(tensor.count(), 0);
-        return StdStringToPyBytes(*tensor.data<string,CPUContext>());
+        return StdStringToPyBytes(*tensor.data<string, CPUContext>());
     }
     }
 };
 };
 
 

Dragon/python/dragon/core/gradient_maker.py

@@ -18,7 +18,6 @@ class GraphGradientMaker(object):
         """ parse ops from string """
         """ parse ops from string """
         g_ops, g_inputs, defaults = CreateGradientDefsCC(op_def.SerializeToString(), g_output)
         g_ops, g_inputs, defaults = CreateGradientDefsCC(op_def.SerializeToString(), g_output)
         for idx, g_op in enumerate(g_ops):
         for idx, g_op in enumerate(g_ops):
-            if sys.version_info >= (3, 0): g_op = g_op.encode()
             new_def = pb.OperatorDef()
             new_def = pb.OperatorDef()
             new_def.ParseFromString(g_op)
             new_def.ParseFromString(g_op)
             _, new_def.name = GetOperatorName()
             _, new_def.name = GetOperatorName()