Fix pybind11 problem

Fix str and bytes problem
Fix sorted problem
Fix math problem
Fix CI problem

paddle/fluid/framework/attribute.h

@@ -82,7 +82,10 @@ class DefaultValueSetter {
  public:
   explicit DefaultValueSetter(T default_value)
       : default_value_(default_value) {}
-  void operator()(T& value) const { value = default_value_; }
+  void operator()(T* value) const {
+    PADDLE_ENFORCE(value != nullptr, "Can not set default value to nullptr");
+    *value = default_value_;
+  }
 
  private:
   T default_value_;
@@ -199,6 +202,7 @@ struct ExtractAttribute<int64_t> {
 template <typename T>
 class TypedAttrChecker {
   typedef std::function<void(T&)> ValueChecker;
+  typedef std::function<void(T*)> ValueSetter;
 
  public:
   explicit TypedAttrChecker(const std::string& attr_name)
@@ -241,7 +245,7 @@ class TypedAttrChecker {
                      "Attribute '%s' is required!", attr_name_);
       // default_value_setter_ has no more than one element
       T val;
-      (default_value_setter_[0])(val);
+      (default_value_setter_[0])(&val);
       attr_map[attr_name_] = val;
     }
     Attribute& attr = attr_map.at(attr_name_);
@@ -255,7 +259,7 @@ class TypedAttrChecker {
  private:
   std::string attr_name_;
   std::vector<ValueChecker> value_checkers_;
-  std::vector<ValueChecker> default_value_setter_;
+  std::vector<ValueSetter> default_value_setter_;
 };
 
 // check whether op's all attributes fit their own limits

paddle/fluid/framework/op_desc.cc

@@ -202,6 +202,57 @@ std::vector<std::string> OpDesc::AttrNames() const {
 }
 
 void OpDesc::SetAttr(const std::string &name, const Attribute &v) {
+  // NOTICE(minqiyang): pybind11 will take the empty list in python as
+  // the std::vector<int> type in C++; so we have to change the attr's type
+  // here if we meet this issue
+  proto::AttrType attr_type = static_cast<proto::AttrType>(v.which() - 1);
+  if (attr_type == proto::AttrType::INTS &&
+      boost::get<std::vector<int>>(v).size() == 0u) {
+    proto::OpProto proto = OpInfoMap::Instance().Get(Type()).Proto();
+    // Find current attr via attr name and set the correct attribute value
+    for (int i = 0; i != proto.attrs_size(); ++i) {
+      const proto::OpProto::Attr &attr = proto.attrs(i);
+      if (attr.name() == name) {
+        switch (attr.type()) {
+          case proto::AttrType::BOOLEANS: {
+            VLOG(11) << "SetAttr: " << Type() << ", " << name
+                     << " from INTS to BOOLEANS";
+            this->attrs_[name] = std::vector<bool>();
+            break;
+          }
+          case proto::AttrType::INTS: {
+            VLOG(11) << "SetAttr: " << Type() << ", " << name
+                     << " from INTS to INTS";
+            this->attrs_[name] = std::vector<int>();
+            break;
+          }
+          case proto::AttrType::FLOATS: {
+            VLOG(11) << "SetAttr: " << Type() << ", " << name
+                     << " from INTS to FLOATS";
+            this->attrs_[name] = std::vector<float>();
+            break;
+          }
+          case proto::AttrType::STRINGS: {
+            VLOG(11) << "SetAttr: " << Type() << ", " << name
+                     << " from INTS to STRINGS";
+            this->attrs_[name] = std::vector<std::string>();
+            break;
+          }
+          case proto::AttrType::BLOCKS: {
+            VLOG(11) << "SetAttr: " << Type() << ", " << name
+                     << " from INTS to BLOCKS";
+            this->SetBlocksAttr(name, std::vector<BlockDesc *>());
+            return;
+          }
+          default:
+            PADDLE_THROW("Wrong attr type %d", attr.type());
+        }
+        need_update_ = true;
+        return;
+      }
+    }
+  }
+
   this->attrs_[name] = v;
   need_update_ = true;
 }

paddle/fluid/pybind/protobuf.cc

@@ -205,11 +205,7 @@ void BindBlockDesc(pybind11::module *m) {
 void BindVarDsec(pybind11::module *m) {
   pybind11::class_<pd::VarDesc> var_desc(*m, "VarDesc", "");
   var_desc
-      .def("name",
-           [](pd::VarDesc &self) {
-             pybind11::bytes name = self.Name();
-             return name;
-           },
+      .def("name", [](pd::VarDesc &self) { return self.Name(); },
            pybind11::return_value_policy::reference)
       .def("set_name", &pd::VarDesc::SetName)
       .def("set_shape", &pd::VarDesc::SetShape)

paddle/fluid/pybind/pybind.cc

@@ -54,6 +54,8 @@ limitations under the License. */
 #include "paddle/fluid/platform/gpu_info.h"
 #endif
 
+#include "pybind11/stl.h"
+
 // disable auto conversion to list in Python
 PYBIND11_MAKE_OPAQUE(paddle::framework::LoDTensorArray);
 

python/paddle/dataset/cifar.py

@@ -53,7 +53,7 @@ def reader_creator(filename, sub_name, cycle=False):
             yield (sample / 255.0).astype(numpy.float32), int(label)
 
     def reader():
-        with tarfile.open(filename, mode='r') as f:
+        with tarfile.open(filename, mode='rb') as f:
             names = (each_item.name for each_item in f
                      if sub_name in each_item.name)
 

python/paddle/dataset/common.py

@@ -20,6 +20,7 @@ import shutil
 import sys
 import importlib
 import paddle.dataset
+import paddle.fluid.compat as cpt
 import six.moves.cPickle as pickle
 import glob
 
@@ -93,7 +94,7 @@ def download(url, module_name, md5sum, save_name=None):
                 total_length = int(total_length)
                 for data in r.iter_content(chunk_size=4096):
                     dl += len(data)
-                    f.write(data)
+                    f.write(cpt.to_literal_str(data))
                     done = int(50 * dl / total_length)
                     sys.stdout.write("\r[%s%s]" % ('=' * done,
                                                    ' ' * (50 - done)))

python/paddle/dataset/image.py

@@ -56,7 +56,7 @@ def batch_images_from_tar(data_file,
     :type data_file: string
     :param dataset_name: 'train','test' or 'valid'
     :type dataset_name: string
-    :param img2label: a dic with image file name as key 
+    :param img2label: a dic with image file name as key
                     and image's label as value
     :type img2label: dic
     :param num_per_batch: image number per batch file
@@ -88,7 +88,7 @@ def batch_images_from_tar(data_file,
                 output['data'] = data
                 pickle.dump(
                     output,
-                    open('%s/batch_%d' % (out_path, file_id), 'w'),
+                    open('%s/batch_%d' % (out_path, file_id), 'wb'),
                     protocol=pickle.HIGHEST_PROTOCOL)
                 file_id += 1
                 data = []
@@ -99,7 +99,7 @@ def batch_images_from_tar(data_file,
         output['data'] = data
         pickle.dump(
             output,
-            open('%s/batch_%d' % (out_path, file_id), 'w'),
+            open('%s/batch_%d' % (out_path, file_id), 'wb'),
             protocol=pickle.HIGHEST_PROTOCOL)
 
     with open(meta_file, 'a') as meta:
@@ -113,7 +113,7 @@ def load_image_bytes(bytes, is_color=True):
     Load an color or gray image from bytes array.
 
     Example usage:
-    
+
     .. code-block:: python
 
         with open('cat.jpg') as f:
@@ -137,7 +137,7 @@ def load_image(file, is_color=True):
     Load an color or gray image from the file path.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = load_image('cat.jpg')
@@ -161,16 +161,16 @@ def load_image(file, is_color=True):
 
 
 def resize_short(im, size):
-    """ 
+    """
     Resize an image so that the length of shorter edge is size.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = load_image('cat.jpg')
         im = resize_short(im, 256)
-    
+
     :param im: the input image with HWC layout.
     :type im: ndarray
     :param size: the shorter edge size of image after resizing.
@@ -193,17 +193,17 @@ def to_chw(im, order=(2, 0, 1)):
     according the order (2,0,1).
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = load_image('cat.jpg')
         im = resize_short(im, 256)
         im = to_chw(im)
-    
+
     :param im: the input image with HWC layout.
     :type im: ndarray
     :param order: the transposed order.
-    :type order: tuple|list 
+    :type order: tuple|list
     """
     assert len(im.shape) == len(order)
     im = im.transpose(order)
@@ -215,11 +215,11 @@ def center_crop(im, size, is_color=True):
     Crop the center of image with size.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = center_crop(im, 224)
-    
+
     :param im: the input image with HWC layout.
     :type im: ndarray
     :param size: the cropping size.
@@ -243,11 +243,11 @@ def random_crop(im, size, is_color=True):
     Randomly crop input image with size.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = random_crop(im, 224)
-    
+
     :param im: the input image with HWC layout.
     :type im: ndarray
     :param size: the cropping size.
@@ -272,11 +272,11 @@ def left_right_flip(im, is_color=True):
     Return the flipped image.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = left_right_flip(im)
-    
+
     :param im: input image with HWC layout or HW layout for gray image
     :type im: ndarray
     :param is_color: whether input image is color or not
@@ -299,7 +299,7 @@ def simple_transform(im,
     resizing, croping and flipping.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = simple_transform(im, 256, 224, True)
@@ -314,7 +314,7 @@ def simple_transform(im,
     :type is_train: bool
     :param is_color: whether the image is color or not.
     :type is_color: bool
-    :param mean: the mean values, which can be element-wise mean values or 
+    :param mean: the mean values, which can be element-wise mean values or
                  mean values per channel.
     :type mean: numpy array | list
     """
@@ -332,7 +332,7 @@ def simple_transform(im,
     im = im.astype('float32')
     if mean is not None:
         mean = np.array(mean, dtype=np.float32)
-        # mean value, may be one value per channel 
+        # mean value, may be one value per channel
         if mean.ndim == 1 and is_color:
             mean = mean[:, np.newaxis, np.newaxis]
         elif mean.ndim == 1:
@@ -357,7 +357,7 @@ def load_and_transform(filename,
     for the transform operations.
 
     Example usage:
-    
+
     .. code-block:: python
 
         im = load_and_transform('cat.jpg', 256, 224, True)
@@ -372,7 +372,7 @@ def load_and_transform(filename,
     :type is_train: bool
     :param is_color: whether the image is color or not.
     :type is_color: bool
-    :param mean: the mean values, which can be element-wise mean values or 
+    :param mean: the mean values, which can be element-wise mean values or
                  mean values per channel.
     :type mean: numpy array | list
     """

python/paddle/dataset/mnist.py

@@ -21,6 +21,8 @@ import paddle.dataset.common
 import subprocess
 import numpy
 import platform
+import six
+import tempfile
 from six.moves import range
 __all__ = ['train', 'test', 'convert']
 
@@ -46,23 +48,28 @@ def reader_creator(image_filename, label_filename, buffer_size):
 
         # According to http://stackoverflow.com/a/38061619/724872, we
         # cannot use standard package gzip here.
-        m = subprocess.Popen([zcat_cmd, image_filename], stdout=subprocess.PIPE)
-        m.stdout.read(16)  # skip some magic bytes
+        tmp_image_file = tempfile.TemporaryFile(prefix='paddle_dataset')
+        m = subprocess.Popen(
+            [zcat_cmd, image_filename], stdout=tmp_image_file).communicate()
+        tmp_image_file.seek(16)  # skip some magic bytes
 
-        l = subprocess.Popen([zcat_cmd, label_filename], stdout=subprocess.PIPE)
-        l.stdout.read(8)  # skip some magic bytes
+        # Python3 will not take stdout as file
+        tmp_label_file = tempfile.TemporaryFile(prefix='paddle_dataset')
+        l = subprocess.Popen(
+            [zcat_cmd, label_filename], stdout=tmp_label_file).communicate()
+        tmp_label_file.seek(8)  # skip some magic bytes
 
         try:  # reader could be break.
             while True:
                 labels = numpy.fromfile(
-                    l.stdout, 'ubyte', count=buffer_size).astype("int")
+                    tmp_label_file, 'ubyte', count=buffer_size).astype("int")
 
                 if labels.size != buffer_size:
                     break  # numpy.fromfile returns empty slice after EOF.
 
                 images = numpy.fromfile(
-                    m.stdout, 'ubyte', count=buffer_size * 28 * 28).reshape(
-                        (buffer_size, 28 * 28)).astype('float32')
+                    tmp_image_file, 'ubyte', count=buffer_size * 28 *
+                    28).reshape((buffer_size, 28 * 28)).astype('float32')
 
                 images = images / 255.0 * 2.0 - 1.0
 

python/paddle/dataset/uci_housing.py

@@ -71,7 +71,7 @@ def load_data(filename, feature_num=14, ratio=0.8):
         return
 
     data = np.fromfile(filename, sep=' ')
-    data = data.reshape(data.shape[0] / feature_num, feature_num)
+    data = data.reshape(data.shape[0] // feature_num, feature_num)
     maximums, minimums, avgs = data.max(axis=0), data.min(axis=0), data.sum(
         axis=0) / data.shape[0]
     feature_range(maximums[:-1], minimums[:-1])

python/paddle/dataset/wmt16.py

@@ -29,6 +29,7 @@ Multi30K: Multilingual English-German Image Descriptions.
 """
 
 import os
+import six
 import tarfile
 import gzip
 from collections import defaultdict
@@ -120,7 +121,7 @@ def reader_creator(tar_file, file_name, src_dict_size, trg_dict_size, src_lang):
 
         with tarfile.open(tar_file, mode="r") as f:
             for line in f.extractfile(file_name):
-                line_split = line.strip().split("\t")
+                line_split = line.strip().split(six.b("\t"))
                 if len(line_split) != 2:
                     continue
                 src_words = line_split[src_col].split()

python/paddle/fluid/backward.py

@@ -17,6 +17,7 @@ from . import core
 import collections
 import copy
 import six
+from . import compat as cpt
 from . import unique_name
 
 __all__ = ['append_backward']
@@ -75,10 +76,10 @@ def _infer_var_data_type_(grad_var_name, block):
     """
     Infer the data type of given grad variable
     """
-    grad_var = block.desc.find_var(grad_var_name.encode("ascii"))
-    fwd_name = _strip_grad_suffix_(grad_var_name.encode("ascii"))
-    if block.desc.has_var_recursive(fwd_name):
-        fwd_var = block.desc.find_var_recursive(fwd_name.encode("ascii"))
+    grad_var = block.desc.find_var(cpt.to_bytes(grad_var_name))
+    fwd_name = _strip_grad_suffix_(grad_var_name)
+    if block.desc.has_var_recursive(cpt.to_bytes(fwd_name)):
+        fwd_var = block.desc.find_var_recursive(cpt.to_bytes(fwd_name))
         grad_var.set_dtype(fwd_var.dtype())
     else:
         grad_var.set_dtype(core.VarDesc.VarType.FP32)
@@ -102,8 +103,10 @@ def _some_in_set_(cands, s):
     """
     if len(cands) == 0:
         return False
-    for c in cands:
-        if c in s:
+    literal_set = cpt.to_literal_str(s)
+    literal_cands = cpt.to_literal_str(cands)
+    for c in literal_cands:
+        if c in literal_set:
             return True
     return False
 
@@ -114,9 +117,8 @@ def _strip_grad_suffix_(name):
     e.g. x@GRAD ==> x
          y@GRAD@RENAME@1 ==> y
     """
-    if isinstance(name, six.text_type):
-        name = name.encode()
-    pos = name.find(six.b(core.grad_var_suffix()))
+    name = cpt.to_literal_str(name)
+    pos = name.find(core.grad_var_suffix())
     return name[:pos] if pos != -1 else name
 
 
@@ -125,9 +127,7 @@ def _append_grad_suffix_(name):
     Append grad suffix to the given variable name
     e.g. x ==> x@GRAD
     """
-    if isinstance(name, six.text_type):
-        name = name.encode()
-    return name + six.b(core.grad_var_suffix())
+    return cpt.to_literal_str(name) + core.grad_var_suffix()
 
 
 def _addup_repetitive_outputs_(op_descs):
@@ -364,7 +364,8 @@ def _append_backward_ops_(block,
 
         # Getting op's corresponding grad_op
         grad_op_desc, op_grad_to_var = core.get_grad_op_desc(
-            op.desc, no_grad_dict[block.idx], grad_sub_block_list)
+            op.desc,
+            cpt.to_literal_str(no_grad_dict[block.idx]), grad_sub_block_list)
 
         grad_op_descs.extend(grad_op_desc)
         grad_to_var.update(op_grad_to_var)
@@ -411,11 +412,10 @@ def _append_backward_vars_(block, start_op_idx, grad_to_var, grad_info_map):
         new_vars = set()
         # create new gradient variables
         for grad_var_name in op_desc.output_arg_names():
-            grad_var_name = grad_var_name.encode("ascii")
-            if block.desc.has_var_recursive(
-                    grad_var_name) or grad_var_name == core.empty_var_name():
+            if block.desc.has_var_recursive(cpt.to_bytes(
+                    grad_var_name)) or grad_var_name == core.empty_var_name():
                 continue
-            block.desc.var(grad_var_name)
+            block.desc.var(cpt.to_bytes(grad_var_name))
             new_vars.add(grad_var_name)
             if grad_var_name not in grad_to_var:
                 continue
@@ -597,11 +597,12 @@ def append_backward(loss, parameter_list=None, no_grad_set=None,
         parameters = parameter_list
     else:
         params = program.global_block().all_parameters()
+        program.global_block().iter_parameters()
         parameters = [param.name for param in params]
 
     params_and_grads = []
     for param in parameters:
-        if param not in grad_info_map:
+        if cpt.to_literal_str(param) not in grad_info_map:
             continue
         grad_info = grad_info_map[param]
         grad_block = grad_info[1]

python/paddle/fluid/compat.py

+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import six
+
+
+#  str and bytes related functions
+def to_literal_str(obj):
+    if isinstance(obj, list):
+        return [_to_literal_str(item) for item in obj]
+    elif isinstance(obj, set):
+        return set([_to_literal_str(item) for item in obj])
+    else:
+        return _to_literal_str(obj)
+
+
+def _to_literal_str(obj):
+    if isinstance(obj, six.binary_type):
+        return obj.decode('latin-1')
+    elif isinstance(obj, six.text_type):
+        return obj
+    else:
+        return six.u(obj)
+
+
+def to_bytes(obj):
+    if isinstance(obj, list):
+        return [_to_bytes(item) for item in obj]
+    elif isinstance(obj, set):
+        return set([_to_bytes(item) for item in obj])
+    else:
+        return _to_bytes(obj)
+
+
+def _to_bytes(obj):
+    if isinstance(obj, six.text_type):
+        return obj.encode('latin-1')
+    elif isinstance(obj, six.binary_type):
+        return obj
+    else:
+        return six.b(obj)
+
+
+# math related functions
+import math
+
+
+def round(x, d=0):
+    """
+    Compatible round which act the same behaviour in Python3.
+
+    Args:
+        x(float) : The number to round halfway.
+
+    Returns:
+        round result of x
+    """
+    p = 10**d
+    return float(math.floor((x * p) + math.copysign(0.5, x))) / p
+
+
+def floor_division(x, y):
+    return x // y

python/paddle/fluid/framework.py

@@ -19,6 +19,7 @@ import six
 
 import numpy as np
 
+from . import compat as cpt
 from .proto import framework_pb2
 try:
     from . import core
@@ -87,7 +88,7 @@ def convert_np_dtype_to_dtype_(np_dtype):
     elif dtype == np.uint8:
         return core.VarDesc.VarType.UINT8
     else:
-        raise ValueError("Not supported numpy dtype " + six.binary_type(dtype))
+        raise ValueError("Not supported numpy dtype %s" % dtype)
 
 
 def dtype_is_floating(dtype):
@@ -198,11 +199,11 @@ class Variable(object):
         if name is None:
             name = unique_name.generate('_generated_var')
         is_new_var = False
-        name = name if isinstance(name, six.binary_type) else name.encode()
-        self.desc = self.block.desc.find_var(name)
+        name = cpt.to_literal_str(name)
+        self.desc = self.block.desc.find_var(cpt.to_bytes(name))
 
         if self.desc is None:
-            self.desc = self.block.desc.var(name)
+            self.desc = self.block.desc.var(cpt.to_bytes(name))
             is_new_var = True
 
         if is_new_var:
@@ -325,7 +326,7 @@ class Variable(object):
 
     @property
     def name(self):
-        return self.desc.name()
+        return cpt.to_literal_str(self.desc.name())
 
     @name.setter
     def name(self, new_name):
@@ -529,10 +530,7 @@ class Operator(object):
                         elif isinstance(arg, six.binary_type):
                             in_arg_names.append(arg.decode())
                         else:
-                            if isinstance(arg.name, six.string_types):
-                                in_arg_names.append(arg.name)
-                            elif isinstance(arg.name, six.binary_type):
-                                in_arg_names.append(arg.name.decode())
+                            in_arg_names.append(cpt.to_literal_str(arg.name))
                     self.desc.set_input(in_proto.name, in_arg_names)
                 else:
                     self.desc.set_input(in_proto.name, [])
@@ -561,12 +559,7 @@ class Operator(object):
                         (out_proto.name, len(out_args)))
                 out_arg_names = []
                 for arg in out_args:
-                    if isinstance(arg.name, six.string_types):
-                        out_arg_names.append(arg.name)
-                    elif isinstance(arg.name, six.binary_type):
-                        out_arg_names.append(arg.name.decode())
-                    else:
-                        out_arg_names.append(six.u(arg.name))
+                    out_arg_names.append(cpt.to_literal_str(arg.name))
                     arg.op = self
                 self.desc.set_output(out_proto.name, out_arg_names)
 
@@ -994,6 +987,9 @@ class Block(object):
         Returns:
             Variable: the Variable with the giving name.
         """
+        name = cpt.to_literal_str(name)
+        new_name = cpt.to_literal_str(new_name)
+
         if not self.has_var(name):
             raise ValueError("var %s is not in current block" % name)
         v = self.var(name)
@@ -1012,9 +1008,9 @@ class Block(object):
         else:
             raise ValueError("unsupported var type: %s", type(v))
         orig_var_type = v.type
-        self.desc._rename_var(name, new_name)
+        self.desc._rename_var(cpt.to_bytes(name), cpt.to_bytes(new_name))
         # NOTE: v is destroyed by C++ after calling _rename_var.
-        d = self.desc.find_var(new_name)
+        d = self.desc.find_var(cpt.to_bytes(new_name))
         if var_type == "Parameter":
             var = Parameter(
                 self,
@@ -1045,7 +1041,7 @@ class Block(object):
 
     def _remove_var(self, name):
         self._sync_with_cpp()
-        self.desc._remove_var(name)
+        self.desc._remove_var(cpt.to_bytes(name))
         del self.vars[name]
 
     def create_parameter(self, *args, **kwargs):
@@ -1128,7 +1124,7 @@ class Block(object):
 
         # sync variables removed from c++ end
         for var in list(self.vars.keys()):
-            if not self.desc.find_var(var):
+            if not self.desc.find_var(cpt.to_bytes(var)):
                 self.vars.pop(var)
 
         # sync operators from cpp

python/paddle/fluid/graphviz.py

@@ -106,7 +106,8 @@ class Graph(object):
     def _rank_repr(self):
         ranks = sorted(
             list(self.rank_groups.items()),
-            cmp=lambda a, b: a[1].priority > b[1].priority)
+            key=functools.cmp_to_key(
+                lambda a, b: a[1].priority > b[1].priority))
         repr = []
         for x in ranks:
             repr.append(str(x[1]))

python/paddle/fluid/io.py

@@ -600,25 +600,15 @@ def save_inference_model(dirname,
             # "./infer_model".
 
     """
-    if isinstance(feeded_var_names, six.binary_type):
+    if isinstance(feeded_var_names, six.string_types):
         feeded_var_names = [feeded_var_names]
-    elif isinstance(feeded_var_names, six.text_type):
-        feeded_var_names = [feeded_var_names.encode()]
     else:
         if len(feeded_var_names) > 0:
             # TODO(paddle-dev): polish these code blocks
             if not (bool(feeded_var_names) and all(
-                    isinstance(name, six.binary_type)
+                    isinstance(name, six.string_types)
                     for name in feeded_var_names)):
-                if not (all(
-                        isinstance(name, six.text_type)
-                        for name in feeded_var_names)):
-                    raise ValueError(
-                        "'feed_var_names' should be a list of str.")
-                else:
-                    feeded_var_names = [
-                        name.encode() for name in feeded_var_names
-                    ]
+                raise ValueError("'feed_var_names' should be a list of str.")
 
     if isinstance(target_vars, Variable):
         target_vars = [target_vars]

python/paddle/fluid/layers/io.py

@@ -751,7 +751,7 @@ def open_files(filenames,
     else:
         buffer_size = int(buffer_size)
 
-    if isinstance(filenames, basestring):
+    if isinstance(filenames, six.string_types):
         filenames = [filenames]
     dtypes = [convert_np_dtype_to_dtype_(dt) for dt in dtypes]
     shape_concat = []

python/paddle/fluid/layers/nn.py

@@ -360,7 +360,7 @@ def dynamic_lstm(input,
     """
 
     helper = LayerHelper('lstm', **locals())
-    size = size / 4
+    size = size // 4
     weight = helper.create_parameter(
         attr=helper.param_attr, shape=[size, 4 * size], dtype=dtype)
     bias_size = [1, 7 * size]
@@ -1498,7 +1498,7 @@ def conv2d(input,
         raise ValueError("use_cudnn should be True or False")
 
     input_shape = input.shape
-    filter_shape = [num_filters, num_filter_channels] + filter_size
+    filter_shape = [num_filters, int(num_filter_channels)] + filter_size
 
     def _get_default_param_initializer():
         std = (2.0 / (filter_size[0]**2 * num_channels))**0.5
@@ -2669,15 +2669,15 @@ def beam_search(pre_ids,
 
     Refer to `Beam search <https://en.wikipedia.org/wiki/Beam_search>`_
     for more details.
-    
-    This layer does the search in beams for one time step. Specifically, it 
+
+    This layer does the search in beams for one time step. Specifically, it
     selects the top-K candidate word ids of current step from :attr:`ids`
     according to their :attr:`scores` for all source sentences, where K is
     :attr:`beam_size` and :attr:`ids, scores` are predicted results from the
     computation cell. Additionally, :attr:`pre_ids` and :attr:`pre_scores` are
     the output of beam_search at previous step, they are needed for special use
     to handle ended candidate translations.
- 
+
     Note that the :attr:`scores` passed in should be accumulated scores, and
     length penalty should be done with extra operators before calculating the
     accumulated scores if needed, also suggest finding top-K before it and
@@ -3878,7 +3878,7 @@ def nce(input,
 def hsigmoid(input, label, num_classes, param_attr=None, bias_attr=None):
     """
     The hierarchical sigmoid operator is used to accelerate the training
-    process of language model. This operator organizes the classes into a 
+    process of language model. This operator organizes the classes into a
     complete binary tree, each leaf node represents a class(a word) and each
     internal node acts as a binary classifier. For each word there's a unique
     path from root to it's leaf node, hsigmoid calculate the cost for each
@@ -3888,9 +3888,9 @@ def hsigmoid(input, label, num_classes, param_attr=None, bias_attr=None):
 
     Refer to `Hierarchical Probabilistic Neural Network Language Model
     <http://www.iro.umontreal.ca/~lisa/pointeurs/hierarchical-nnlm-aistats05.pdf>`_
-    
+
     Args:
-        input (Variable): The input tensor variable with shape 
+        input (Variable): The input tensor variable with shape
             :math:`[N \\times D]`, where :math:`N` is the size of mini-batch,
             and :math:`D` is the feature size.
         label (Variable): The tensor variable contains labels of training data.
@@ -3898,7 +3898,7 @@ def hsigmoid(input, label, num_classes, param_attr=None, bias_attr=None):
         num_classes: (int), The number of classes, must not be less than 2.
         param_attr (ParamAttr|list of ParamAttr, default None): The parameter
              attribute for learnable parameters/weights of this layer.
-        bias_attr (ParamAttr|list of ParamAttr, default None):  The parameter 
+        bias_attr (ParamAttr|list of ParamAttr, default None):  The parameter
              attribute for the bias of this layer. If it is set to False, no
              bias will be applied.
 
@@ -5293,23 +5293,23 @@ def rank_loss(label, left, right, name=None):
     is a pairwise ranking model with a training sample consisting of a pair
     of documents, A and B. Label P indicates whether A is ranked higher than B
     or not:
- 
+
     P = {0, 1} or {0, 0.5, 1}, where 0.5 means that there is no information
     about the rank of the input pair.
-    
+
     Rank loss layer takes three inputs: left (o_i), right (o_j) and
     label (P_{i,j}). The inputs respectively represent RankNet's output scores
     for documents A and B and the value of label P. The following equation
     computes rank loss C_{i,j} from the inputs:
-    
+
     $$
       C_{i,j} = -\tilde{P_{ij}} * o_{i,j} + \log(1 + e^{o_{i,j}}) \\
       o_{i,j} =  o_i - o_j  \\
       \tilde{P_{i,j}} = \left \{0, 0.5, 1 \right \} \ or \ \left \{0, 1 \right \}
     $$
-    
-    Rank loss layer takes batch inputs with size batch_size (batch_size >= 1).   
- 
+
+    Rank loss layer takes batch inputs with size batch_size (batch_size >= 1).
+
     Args:
         label (Variable): Indicats whether A ranked higher than B or not.
         left (Variable): RankNet's output score for doc A.

python/paddle/fluid/parallel_executor.py

@@ -17,6 +17,7 @@ import multiprocessing
 from . import core
 from . import framework
 from . import executor
+from . import compat as cpt
 import warnings
 import sys
 import six
@@ -154,11 +155,14 @@ class ParallelExecutor(object):
         self.executor = core.ParallelExecutor(
             self._places,
             set([
-                p.name for p in main.global_block().iter_parameters()
+                cpt.to_literal_str(p.name)
+                for p in main.global_block().iter_parameters()
                 if not p.stop_gradient
             ]),
-            set(self.persistable_vars), main.desc, loss_name
-            if loss_name else '', scope, local_scopes, exec_strategy,
+            set(cpt.to_literal_str(var)
+                for var in self.persistable_vars), main.desc,
+            cpt.to_literal_str(loss_name)
+            if loss_name else six.u(''), scope, local_scopes, exec_strategy,
             build_strategy, num_trainers, trainer_id)
         self.scope = scope
 
@@ -270,7 +274,8 @@ class ParallelExecutor(object):
             self.executor.feed_tensors_into_local_scopes(res)
 
         fetch_var_name = '@FETCHED_VAR_NAME@'
-        self.executor.run(fetch_list, fetch_var_name)
+        self.executor.run(
+            cpt.to_literal_str(fetch_list), cpt.to_literal_str(fetch_var_name))
         arr = self.scope.find_var(fetch_var_name).get_lod_tensor_array()
 
         if self.is_dist:

python/paddle/fluid/tests/book/high-level-api/image_classification/cifar10_small_test_set.py

@@ -30,7 +30,7 @@ images per class.
 
 import itertools
 import numpy
-import paddle.v2.dataset.common
+import paddle.dataset.common
 import tarfile
 from six.moves import cPickle as pickle
 from six.moves import zip
@@ -78,6 +78,6 @@ def train10(batch_size=None):
     :rtype: callable
     """
     return reader_creator(
-        paddle.v2.dataset.common.download(CIFAR10_URL, 'cifar', CIFAR10_MD5),
+        paddle.dataset.common.download(CIFAR10_URL, 'cifar', CIFAR10_MD5),
         'data_batch',
         batch_size=batch_size)

python/paddle/fluid/tests/book/test_image_classification.py

@@ -60,7 +60,7 @@ def resnet_cifar10(input, depth=32):
         return tmp
 
     assert (depth - 2) % 6 == 0
-    n = (depth - 2) / 6
+    n = (depth - 2) // 6
     conv1 = conv_bn_layer(
         input=input, ch_out=16, filter_size=3, stride=1, padding=1)
     res1 = layer_warp(basicblock, conv1, 16, 16, n, 1)

python/paddle/fluid/tests/unittests/op_test.py

@@ -15,6 +15,7 @@
 import unittest
 import numpy as np
 import random
+import six
 import time
 import itertools
 import collections
@@ -26,15 +27,13 @@ from paddle.fluid.op import Operator
 from paddle.fluid.executor import Executor
 from paddle.fluid.framework import Program, OpProtoHolder, Variable
 from testsuite import create_op, set_input, append_input_output, append_loss_ops
-from functools import reduce
-from six.moves import zip
 
 
 def randomize_probability(batch_size, class_num, dtype='float32'):
     prob = np.random.uniform(
         0.1, 1.0, size=(batch_size, class_num)).astype(dtype)
     prob_sum = prob.sum(axis=1)
-    for i in range(len(prob)):
+    for i in six.moves.xrange(len(prob)):
         prob[i] /= prob_sum[i]
     return prob
 
@@ -51,7 +50,7 @@ def get_numeric_gradient(place,
     set_input(scope, op, inputs, place)
 
     def product(dim):
-        return reduce(lambda a, b: a * b, dim, 1)
+        return six.moves.reduce(lambda a, b: a * b, dim, 1)
 
     def get_output():
         sum = []
@@ -103,7 +102,7 @@ def get_numeric_gradient(place,
 
     # we only compute gradient of one element each time.
     # we use a for loop to compute the gradient of every element.
-    for i in range(tensor_size):
+    for i in six.moves.xrange(tensor_size):
         if in_place:
             set_input(scope, op, inputs, place)
 
@@ -161,7 +160,7 @@ class OpTest(unittest.TestCase):
             assert isinstance(
                 numpy_dict,
                 dict), "self.inputs, self.outputs must be numpy_dict"
-            for var_name, var_value in numpy_dict.items():
+            for var_name, var_value in six.iteritems(numpy_dict):
                 if isinstance(var_value, (np.ndarray, np.generic)):
                     self.try_call_once(var_value.dtype)
                 elif isinstance(var_value, (list, tuple)):
@@ -225,7 +224,7 @@ class OpTest(unittest.TestCase):
 
     def _get_io_vars(self, block, numpy_inputs):
         inputs = {}
-        for name, value in numpy_inputs.items():
+        for name, value in six.iteritems(numpy_inputs):
             if isinstance(value, list):
                 var_list = [
                     block.var(sub_name) for sub_name, sub_value in value
@@ -268,7 +267,7 @@ class OpTest(unittest.TestCase):
         # if the fetch_list is customized by user, we use it directly.
         # if not, fill the fetch_list by the user configured outputs in test.
         if len(fetch_list) == 0:
-            for var_name, var in outputs.items():
+            for var_name, var in six.iteritems(outputs):
                 if isinstance(var, list):
                     for v in var:
                         fetch_list.append(v)
@@ -371,7 +370,7 @@ class OpTest(unittest.TestCase):
     def __assert_is_close(self, numeric_grads, analytic_grads, names,
                           max_relative_error, msg_prefix):
 
-        for a, b, name in zip(numeric_grads, analytic_grads, names):
+        for a, b, name in six.moves.zip(numeric_grads, analytic_grads, names):
             abs_a = np.abs(a)
             abs_a[abs_a < 1e-3] = 1
 

python/paddle/fluid/tests/unittests/test_data_balance.py

@@ -14,7 +14,7 @@
 
 import unittest
 import paddle.fluid as fluid
-import paddle.v2 as paddle
+import paddle as paddle
 import numpy as np
 
 

python/paddle/fluid/tests/unittests/test_dist_base.py

@@ -16,9 +16,12 @@ import time
 import unittest
 import os
 import sys
+import six
 import signal
 import subprocess
 
+import paddle.fluid.compat as cpt
+
 
 class TestDistBase(unittest.TestCase):
     def setUp(self):
@@ -78,7 +81,7 @@ class TestDistBase(unittest.TestCase):
             env=env_local)
         local_proc.wait()
         out, err = local_proc.communicate()
-        local_ret = out
+        local_ret = cpt.to_literal_str(out)
         sys.stderr.write('local_loss: %s\n' % local_ret)
         sys.stderr.write('local_stderr: %s\n' % err)
 
@@ -116,7 +119,7 @@ class TestDistBase(unittest.TestCase):
         tr1_proc.wait()
         out, err = tr0_proc.communicate()
         sys.stderr.write('dist_stderr: %s\n' % err)
-        loss_data0 = out
+        loss_data0 = cpt.to_literal_str(out)
         sys.stderr.write('dist_loss: %s\n' % loss_data0)
         lines = loss_data0.split("\n")
         dist_first_loss = eval(lines[0].replace(" ", ","))[0]

python/paddle/fluid/tests/unittests/test_pool2d_op.py

@@ -29,11 +29,11 @@ def max_pool2D_forward_naive(x,
     if global_pool == 1:
         ksize = [H, W]
     H_out = (H - ksize[0] + 2 * paddings[0] + strides[0] - 1
-             ) / strides[0] + 1 if ceil_mode else (H - ksize[0] + 2 *
-                                                   paddings[0]) / strides[0] + 1
+             ) // strides[0] + 1 if ceil_mode else (
+                 H - ksize[0] + 2 * paddings[0]) // strides[0] + 1
     W_out = (W - ksize[1] + 2 * paddings[1] + strides[1] - 1
-             ) / strides[1] + 1 if ceil_mode else (W - ksize[1] + 2 *
-                                                   paddings[1]) / strides[1] + 1
+             ) // strides[1] + 1 if ceil_mode else (
+                 W - ksize[1] + 2 * paddings[1]) // strides[1] + 1
     out = np.zeros((N, C, H_out, W_out))
     for i in range(H_out):
         for j in range(W_out):
@@ -57,11 +57,11 @@ def avg_pool2D_forward_naive(x,
     if global_pool == 1:
         ksize = [H, W]
     H_out = (H - ksize[0] + 2 * paddings[0] + strides[0] - 1
-             ) / strides[0] + 1 if ceil_mode else (H - ksize[0] + 2 *
-                                                   paddings[0]) / strides[0] + 1
+             ) // strides[0] + 1 if ceil_mode else (
+                 H - ksize[0] + 2 * paddings[0]) // strides[0] + 1
     W_out = (W - ksize[1] + 2 * paddings[1] + strides[1] - 1
-             ) / strides[1] + 1 if ceil_mode else (W - ksize[1] + 2 *
-                                                   paddings[1]) / strides[1] + 1
+             ) // strides[1] + 1 if ceil_mode else (
+                 W - ksize[1] + 2 * paddings[1]) // strides[1] + 1
     out = np.zeros((N, C, H_out, W_out))
     for i in range(H_out):
         for j in range(W_out):

python/paddle/fluid/tests/unittests/test_reader_reset.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import paddle.fluid as fluid
-import paddle.v2 as paddle
+import paddle as paddle
 import numpy as np
 import unittest
 

python/paddle/fluid/tests/unittests/test_reorder_lod_tensor.py

@@ -17,6 +17,7 @@ import paddle.fluid as fluid
 import paddle.fluid.core as core
 from paddle.fluid.layers.control_flow import lod_rank_table
 import numpy
+import functools
 
 
 class TestReorderLoDTensor(unittest.TestCase):
@@ -101,7 +102,8 @@ class TestReorderLoDTensor(unittest.TestCase):
         rank_table = []  # list of (index, length)
         for i in range(len(ref_lod)):
             rank_table.append((i, ref_lod[i]))
-        rank_table = sorted(rank_table, lambda x, y: y[1] - x[1])
+        rank_table = sorted(
+            rank_table, key=functools.cmp_to_key(lambda x, y: y[1] - x[1]))
 
         # compute the input sequence info according to input_lod
         input_value, input_lod = self.data[self.data_desc[0][0]]

python/paddle/fluid/tests/unittests/test_roi_pool_op.py

@@ -16,6 +16,7 @@ import unittest
 import numpy as np
 import math
 import sys
+import paddle.fluid.compat as cpt
 from op_test import OpTest
 
 
@@ -59,10 +60,10 @@ class TestROIPoolOp(OpTest):
         for i in range(self.rois_num):
             roi = self.rois[i]
             roi_batch_id = roi[0]
-            roi_start_w = int(round(roi[1] * self.spatial_scale))
-            roi_start_h = int(round(roi[2] * self.spatial_scale))
-            roi_end_w = int(round(roi[3] * self.spatial_scale))
-            roi_end_h = int(round(roi[4] * self.spatial_scale))
+            roi_start_w = int(cpt.round(roi[1] * self.spatial_scale))
+            roi_start_h = int(cpt.round(roi[2] * self.spatial_scale))
+            roi_end_w = int(cpt.round(roi[3] * self.spatial_scale))
+            roi_end_h = int(cpt.round(roi[4] * self.spatial_scale))
 
             roi_height = int(max(roi_end_h - roi_start_h + 1, 1))
             roi_width = int(max(roi_end_w - roi_start_w + 1, 1))
@@ -97,8 +98,8 @@ class TestROIPoolOp(OpTest):
                             for w in range(wstart, wend):
                                 if x_i[c, h, w] > out_data[i, c, ph, pw]:
                                     out_data[i, c, ph, pw] = x_i[c, h, w]
-                                    argmax_data[i, c, ph, pw] = h * \
-                                        self.width + w
+                                    argmax_data[i, c, ph,
+                                                pw] = h * self.width + w
 
         self.outs = out_data.astype('float32')
         self.argmaxes = argmax_data.astype('int64')
@@ -110,14 +111,14 @@ class TestROIPoolOp(OpTest):
             self.rois_lod[0].append(bno + 1)
             for i in range(bno + 1):
                 x1 = np.random.random_integers(
-                    0, self.width / self.spatial_scale - self.pooled_width)
+                    0, self.width // self.spatial_scale - self.pooled_width)
                 y1 = np.random.random_integers(
-                    0, self.height / self.spatial_scale - self.pooled_height)
+                    0, self.height // self.spatial_scale - self.pooled_height)
 
                 x2 = np.random.random_integers(x1 + self.pooled_width,
-                                               self.width / self.spatial_scale)
-                y2 = np.random.random_integers(y1 + self.pooled_height,
-                                               self.height / self.spatial_scale)
+                                               self.width // self.spatial_scale)
+                y2 = np.random.random_integers(
+                    y1 + self.pooled_height, self.height // self.spatial_scale)
 
                 roi = [bno, x1, y1, x2, y2]
                 rois.append(roi)

python/paddle/fluid/tests/unittests/test_unpool_op.py

@@ -27,7 +27,7 @@ def unpool2dmax_forward_naive(input, indices, ksize, strides, paddings):
             for h in range(s2):
                 for w in range(s3):
                     index = indices[nidx, cidx, h, w]
-                    hidx = (index - index % out_wsize) / out_wsize
+                    hidx = (index - index % out_wsize) // out_wsize
                     widx = index % out_wsize
                     out[nidx, cidx, int(hidx), int(widx)] = \
                             input[nidx, cidx, h, w]
@@ -41,9 +41,9 @@ class TestUnpoolOp(OpTest):
         self.init_test_case()
         pre_input = np.random.random(self.shape).astype("float32")
         nsize, csize, hsize, wsize = pre_input.shape
-        hsize_out = (hsize - self.ksize[0] + 2 * self.paddings[0]) / \
+        hsize_out = (hsize - self.ksize[0] + 2 * self.paddings[0]) // \
                 self.strides[0] + 1
-        wsize_out = (wsize - self.ksize[1] + 2 * self.paddings[1]) / \
+        wsize_out = (wsize - self.ksize[1] + 2 * self.paddings[1]) // \
                 self.strides[1] + 1
         input = np.zeros((nsize, csize, hsize_out, wsize_out))
         indices = np.zeros((nsize, csize, hsize_out, wsize_out))
@@ -62,7 +62,7 @@ class TestUnpoolOp(OpTest):
                         input[nidx, cidx, i, j] = x_masked.max()
                         arg = x_masked.argmax()
                         indices[nidx, cidx, i, j] = \
-                                (r_start + arg / self.ksize[1]) * wsize + \
+                                (r_start + arg // self.ksize[1]) * wsize + \
                                 c_start + arg % self.ksize[1]
         output = self.unpool2d_forward_naive(input, indices, self.ksize, \
                 self.strides, self.paddings).astype("float32")

python/paddle/fluid/tests/unittests/test_warpctc_op.py

@@ -132,7 +132,7 @@ class CTCForward(object):
             for k in range(end - start):
                 j = k + start
                 if j & 1 == 1:
-                    label_idx = j / 2
+                    label_idx = j // 2
                     label_val = labels_a_sequence[label_idx, 0]
                     fv = self.log_add(forward_vars[i - 1, j],
                                       forward_vars[i - 1, j - 1])

python/paddle/fluid/transpiler/details/program_utils.py

@@ -12,12 +12,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import six
+
 
 def delete_ops(block, ops):
     try:
         start = list(block.ops).index(ops[0])
         end = list(block.ops).index(ops[-1])
-        [block._remove_op(start) for _ in range(end - start + 1)]
+        [block._remove_op(start) for _ in six.moves.range(end - start + 1)]
     except Exception as e:
         raise e
     block.program._sync_with_cpp()

python/paddle/fluid/transpiler/distribute_transpiler.py

@@ -1017,7 +1017,7 @@ class DistributeTranspiler(object):
 
             for i, block in enumerate(splited):
                 size = block[1]
-                rows = size / orig_dim1_flatten
+                rows = size // orig_dim1_flatten
                 splited_shape = [rows]
                 if len(orig_shape) >= 2:
                     splited_shape.extend(orig_shape[1:])

tools/test_runner.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 from __future__ import print_function
+
 import unittest
 import os
 import sys