From e2afe5e054e1c5f7e294dd3b7e51453be8e2d741 Mon Sep 17 00:00:00 2001
From: Thiago Crepaldi <thiago.crepaldi@microsoft.com>
Date: Thu, 2 Apr 2020 03:13:46 +0000
Subject: [PATCH] Revert Session and InferenceSession implementation

---
 onnxruntime/core/session/inference_session.cc |  3 +-
 onnxruntime/python/session.py                 | 80 +++++++++----------
 2 files changed, 39 insertions(+), 44 deletions(-)

diff --git a/onnxruntime/core/session/inference_session.cc b/onnxruntime/core/session/inference_session.cc
index 852fb46bd8..9d1ba34ef7 100644
--- a/onnxruntime/core/session/inference_session.cc
+++ b/onnxruntime/core/session/inference_session.cc
@@ -1155,7 +1155,8 @@ Status InferenceSession::Run(const RunOptions& run_options, const std::vector<st
     }
     // execute the graph
     ORT_CHECK_AND_SET_RETVAL(utils::ExecuteGraph(*session_state_, feeds_fetches_manager, feeds, *p_fetches,
-                                                 session_options_.execution_mode, run_options.terminate, run_logger));
+                                                 session_options_.execution_mode, run_options.terminate, run_logger,
+                                                 run_options.only_execute_path_to_fetches));
 
   } catch (const std::exception& e) {
     retval = Status(common::ONNXRUNTIME, common::FAIL, e.what());
diff --git a/onnxruntime/python/session.py b/onnxruntime/python/session.py
index 3ca173bed5..f17736ee0a 100644
--- a/onnxruntime/python/session.py
+++ b/onnxruntime/python/session.py
@@ -20,49 +20,9 @@ class Session:
     """
     This is the main class used to run a model.
     """
-
-    def __init__(self, path_or_bytes, sess_options=None, providers=[]):
-        """
-        :param path_or_bytes: filename or serialized model in a byte string
-        :param sess_options: session options
-        :param providers: providers to use for session. If empty, will use
-            all available providers.
-        """
-        self._path_or_bytes = path_or_bytes
-        self._sess_options = sess_options
-        self._load_model(providers)
+    def __init__(self, sess):
         self._enable_fallback = True
 
-    def _load_model(self, providers=[]):
-        if isinstance(self._path_or_bytes, str):
-            self._sess = C.InferenceSession(
-                self._sess_options if self._sess_options else C.get_default_session_options(), self._path_or_bytes,
-                True)
-        elif isinstance(self._path_or_bytes, bytes):
-            self._sess = C.InferenceSession(
-                self._sess_options if self._sess_options else C.get_default_session_options(), self._path_or_bytes,
-                False)
-        # elif isinstance(self._path_or_bytes, tuple):
-        # to remove, hidden trick
-        #   self._sess.load_model_no_init(self._path_or_bytes[0], providers)
-        else:
-            raise TypeError("Unable to load from type '{0}'".format(type(self._path_or_bytes)))
-
-        self._sess.load_model(providers)
-
-        self._session_options = self._sess.session_options
-        self._inputs_meta = self._sess.inputs_meta
-        self._outputs_meta = self._sess.outputs_meta
-        self._overridable_initializers = self._sess.overridable_initializers
-        self._model_meta = self._sess.model_meta
-        self._providers = self._sess.get_providers()
-
-        # Tensorrt can fall back to CUDA. All others fall back to CPU.
-        if 'TensorrtExecutionProvider' in C.get_available_providers():
-            self._fallback_providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
-        else:
-            self._fallback_providers = ['CPUExecutionProvider']
-
     def _reset_session(self):
         "release underlying session object."
         # meta data references session internal structures
@@ -76,7 +36,7 @@ class Session:
 
     def get_session_options(self):
         "Return the session options. See :class:`onnxruntime.SessionOptions`."
-        return self._session_options
+        return self._sess_options
 
     def get_inputs(self):
         "Return the inputs metadata as a list of :class:`onnxruntime.NodeArg`."
@@ -193,7 +153,41 @@ class InferenceSession(Session):
         :param providers: providers to use for session. If empty, will use
             all available providers.
         """
-        Session.__init__(self, path_or_bytes, sess_options, providers)
+        self._path_or_bytes = path_or_bytes
+        self._sess_options = sess_options
+        self._load_model(providers)
+        self._enable_fallback = True
+        Session.__init__(self, self._sess)
+
+    def _load_model(self, providers=[]):
+        if isinstance(self._path_or_bytes, str):
+            self._sess = C.InferenceSession(
+                self._sess_options if self._sess_options else C.get_default_session_options(), self._path_or_bytes,
+                True)
+        elif isinstance(self._path_or_bytes, bytes):
+            self._sess = C.InferenceSession(
+                self._sess_options if self._sess_options else C.get_default_session_options(), self._path_or_bytes,
+                False)
+        # elif isinstance(self._path_or_bytes, tuple):
+        # to remove, hidden trick
+        #   self._sess.load_model_no_init(self._path_or_bytes[0], providers)
+        else:
+            raise TypeError("Unable to load from type '{0}'".format(type(self._path_or_bytes)))
+
+        self._sess.load_model(providers)
+
+        self._sess_options = self._sess.session_options
+        self._inputs_meta = self._sess.inputs_meta
+        self._outputs_meta = self._sess.outputs_meta
+        self._overridable_initializers = self._sess.overridable_initializers
+        self._model_meta = self._sess.model_meta
+        self._providers = self._sess.get_providers()
+
+        # Tensorrt can fall back to CUDA. All others fall back to CPU.
+        if 'TensorrtExecutionProvider' in C.get_available_providers():
+            self._fallback_providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+        else:
+            self._fallback_providers = ['CPUExecutionProvider']
 
 
 class IOBinding: