Add protobuf 2.2.0a sources

This is the contents of protobuf-2.2.0a.tar.bz2 from
http://code.google.com/p/protobuf/downloads/list and
is the base code for the javamicro code generator.

Change-Id: Ie9a0440a824d615086445b6636944484b3155afa

23 files changed, 6565 insertions, 0 deletions

diff --git a/python/google/protobuf/internal/__init__.py b/python/google/protobuf/internal/__init__.py
new file mode 100755
index 0000000..e69de29
--- /dev/null
+++ b/python/google/protobuf/internal/__init__.py
diff --git a/python/google/protobuf/internal/containers.py b/python/google/protobuf/internal/containers.py
new file mode 100755
index 0000000..d8a825d
--- /dev/null
+++ b/python/google/protobuf/internal/containers.py
@@ -0,0 +1,227 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Contains container classes to represent different protocol buffer types.
+
+This file defines container classes which represent categories of protocol
+buffer field types which need extra maintenance. Currently these categories
+are:
+  - Repeated scalar fields - These are all repeated fields which aren't
+    composite (e.g. they are of simple types like int32, string, etc).
+  - Repeated composite fields - Repeated fields which are composite. This
+    includes groups and nested messages.
+"""
+
+__author__ = 'petar@google.com (Petar Petrov)'
+
+
+class BaseContainer(object):
+
+  """Base container class."""
+
+  # Minimizes memory usage and disallows assignment to other attributes.
+  __slots__ = ['_message_listener', '_values']
+
+  def __init__(self, message_listener):
+    """
+    Args:
+      message_listener: A MessageListener implementation.
+        The RepeatedScalarFieldContainer will call this object's
+        TransitionToNonempty() method when it transitions from being empty to
+        being nonempty.
+    """
+    self._message_listener = message_listener
+    self._values = []
+
+  def __getitem__(self, key):
+    """Retrieves item by the specified key."""
+    return self._values[key]
+
+  def __len__(self):
+    """Returns the number of elements in the container."""
+    return len(self._values)
+
+  def __ne__(self, other):
+    """Checks if another instance isn't equal to this one."""
+    # The concrete classes should define __eq__.
+    return not self == other
+
+
+class RepeatedScalarFieldContainer(BaseContainer):
+
+  """Simple, type-checked, list-like container for holding repeated scalars."""
+
+  # Disallows assignment to other attributes.
+  __slots__ = ['_type_checker']
+
+  def __init__(self, message_listener, type_checker):
+    """
+    Args:
+      message_listener: A MessageListener implementation.
+        The RepeatedScalarFieldContainer will call this object's
+        TransitionToNonempty() method when it transitions from being empty to
+        being nonempty.
+      type_checker: A type_checkers.ValueChecker instance to run on elements
+        inserted into this container.
+    """
+    super(RepeatedScalarFieldContainer, self).__init__(message_listener)
+    self._type_checker = type_checker
+
+  def append(self, value):
+    """Appends an item to the list. Similar to list.append()."""
+    self.insert(len(self._values), value)
+
+  def insert(self, key, value):
+    """Inserts the item at the specified position. Similar to list.insert()."""
+    self._type_checker.CheckValue(value)
+    self._values.insert(key, value)
+    self._message_listener.ByteSizeDirty()
+    if len(self._values) == 1:
+      self._message_listener.TransitionToNonempty()
+
+  def extend(self, elem_seq):
+    """Extends by appending the given sequence. Similar to list.extend()."""
+    if not elem_seq:
+      return
+
+    orig_empty = len(self._values) == 0
+    new_values = []
+    for elem in elem_seq:
+      self._type_checker.CheckValue(elem)
+      new_values.append(elem)
+    self._values.extend(new_values)
+    self._message_listener.ByteSizeDirty()
+    if orig_empty:
+      self._message_listener.TransitionToNonempty()
+
+  def remove(self, elem):
+    """Removes an item from the list. Similar to list.remove()."""
+    self._values.remove(elem)
+    self._message_listener.ByteSizeDirty()
+
+  def __setitem__(self, key, value):
+    """Sets the item on the specified position."""
+    # No need to call TransitionToNonempty(), since if we're able to
+    # set the element at this index, we were already nonempty before
+    # this method was called.
+    self._message_listener.ByteSizeDirty()
+    self._type_checker.CheckValue(value)
+    self._values[key] = value
+
+  def __getslice__(self, start, stop):
+    """Retrieves the subset of items from between the specified indices."""
+    return self._values[start:stop]
+
+  def __setslice__(self, start, stop, values):
+    """Sets the subset of items from between the specified indices."""
+    new_values = []
+    for value in values:
+      self._type_checker.CheckValue(value)
+      new_values.append(value)
+    self._values[start:stop] = new_values
+    self._message_listener.ByteSizeDirty()
+
+  def __delitem__(self, key):
+    """Deletes the item at the specified position."""
+    del self._values[key]
+    self._message_listener.ByteSizeDirty()
+
+  def __delslice__(self, start, stop):
+    """Deletes the subset of items from between the specified indices."""
+    del self._values[start:stop]
+    self._message_listener.ByteSizeDirty()
+
+  def __eq__(self, other):
+    """Compares the current instance with another one."""
+    if self is other:
+      return True
+    # Special case for the same type which should be common and fast.
+    if isinstance(other, self.__class__):
+      return other._values == self._values
+    # We are presumably comparing against some other sequence type.
+    return other == self._values
+
+
+class RepeatedCompositeFieldContainer(BaseContainer):
+
+  """Simple, list-like container for holding repeated composite fields."""
+
+  # Disallows assignment to other attributes.
+  __slots__ = ['_message_descriptor']
+
+  def __init__(self, message_listener, message_descriptor):
+    """
+    Note that we pass in a descriptor instead of the generated directly,
+    since at the time we construct a _RepeatedCompositeFieldContainer we
+    haven't yet necessarily initialized the type that will be contained in the
+    container.
+
+    Args:
+      message_listener: A MessageListener implementation.
+        The RepeatedCompositeFieldContainer will call this object's
+        TransitionToNonempty() method when it transitions from being empty to
+        being nonempty.
+      message_descriptor: A Descriptor instance describing the protocol type
+        that should be present in this container.  We'll use the
+        _concrete_class field of this descriptor when the client calls add().
+    """
+    super(RepeatedCompositeFieldContainer, self).__init__(message_listener)
+    self._message_descriptor = message_descriptor
+
+  def add(self):
+    new_element = self._message_descriptor._concrete_class()
+    new_element._SetListener(self._message_listener)
+    self._values.append(new_element)
+    self._message_listener.ByteSizeDirty()
+    self._message_listener.TransitionToNonempty()
+    return new_element
+
+  def __getslice__(self, start, stop):
+    """Retrieves the subset of items from between the specified indices."""
+    return self._values[start:stop]
+
+  def __delitem__(self, key):
+    """Deletes the item at the specified position."""
+    del self._values[key]
+    self._message_listener.ByteSizeDirty()
+
+  def __delslice__(self, start, stop):
+    """Deletes the subset of items from between the specified indices."""
+    del self._values[start:stop]
+    self._message_listener.ByteSizeDirty()
+
+  def __eq__(self, other):
+    """Compares the current instance with another one."""
+    if self is other:
+      return True
+    if not isinstance(other, self.__class__):
+      raise TypeError('Can only compare repeated composite fields against '
+                      'other repeated composite fields.')
+    return self._values == other._values
diff --git a/python/google/protobuf/internal/decoder.py b/python/google/protobuf/internal/decoder.py
new file mode 100755
index 0000000..83d6fe0
--- /dev/null
+++ b/python/google/protobuf/internal/decoder.py
@@ -0,0 +1,209 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Class for decoding protocol buffer primitives.
+
+Contains the logic for decoding every logical protocol field type
+from one of the 5 physical wire types.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import struct
+from google.protobuf import message
+from google.protobuf.internal import input_stream
+from google.protobuf.internal import wire_format
+
+
+
+# Note that much of this code is ported from //net/proto/ProtocolBuffer, and
+# that the interface is strongly inspired by WireFormat from the C++ proto2
+# implementation.
+
+
+class Decoder(object):
+
+  """Decodes logical protocol buffer fields from the wire."""
+
+  def __init__(self, s):
+    """Initializes the decoder to read from s.
+
+    Args:
+      s: An immutable sequence of bytes, which must be accessible
+        via the Python buffer() primitive (i.e., buffer(s)).
+    """
+    self._stream = input_stream.InputStream(s)
+
+  def EndOfStream(self):
+    """Returns true iff we've reached the end of the bytes we're reading."""
+    return self._stream.EndOfStream()
+
+  def Position(self):
+    """Returns the 0-indexed position in |s|."""
+    return self._stream.Position()
+
+  def ReadFieldNumberAndWireType(self):
+    """Reads a tag from the wire. Returns a (field_number, wire_type) pair."""
+    tag_and_type = self.ReadUInt32()
+    return wire_format.UnpackTag(tag_and_type)
+
+  def SkipBytes(self, bytes):
+    """Skips the specified number of bytes on the wire."""
+    self._stream.SkipBytes(bytes)
+
+  # Note that the Read*() methods below are not exactly symmetrical with the
+  # corresponding Encoder.Append*() methods.  Those Encoder methods first
+  # encode a tag, but the Read*() methods below assume that the tag has already
+  # been read, and that the client wishes to read a field of the specified type
+  # starting at the current position.
+
+  def ReadInt32(self):
+    """Reads and returns a signed, varint-encoded, 32-bit integer."""
+    return self._stream.ReadVarint32()
+
+  def ReadInt64(self):
+    """Reads and returns a signed, varint-encoded, 64-bit integer."""
+    return self._stream.ReadVarint64()
+
+  def ReadUInt32(self):
+    """Reads and returns an signed, varint-encoded, 32-bit integer."""
+    return self._stream.ReadVarUInt32()
+
+  def ReadUInt64(self):
+    """Reads and returns an signed, varint-encoded,64-bit integer."""
+    return self._stream.ReadVarUInt64()
+
+  def ReadSInt32(self):
+    """Reads and returns a signed, zigzag-encoded, varint-encoded,
+    32-bit integer."""
+    return wire_format.ZigZagDecode(self._stream.ReadVarUInt32())
+
+  def ReadSInt64(self):
+    """Reads and returns a signed, zigzag-encoded, varint-encoded,
+    64-bit integer."""
+    return wire_format.ZigZagDecode(self._stream.ReadVarUInt64())
+
+  def ReadFixed32(self):
+    """Reads and returns an unsigned, fixed-width, 32-bit integer."""
+    return self._stream.ReadLittleEndian32()
+
+  def ReadFixed64(self):
+    """Reads and returns an unsigned, fixed-width, 64-bit integer."""
+    return self._stream.ReadLittleEndian64()
+
+  def ReadSFixed32(self):
+    """Reads and returns a signed, fixed-width, 32-bit integer."""
+    value = self._stream.ReadLittleEndian32()
+    if value >= (1 << 31):
+      value -= (1 << 32)
+    return value
+
+  def ReadSFixed64(self):
+    """Reads and returns a signed, fixed-width, 64-bit integer."""
+    value = self._stream.ReadLittleEndian64()
+    if value >= (1 << 63):
+      value -= (1 << 64)
+    return value
+
+  def ReadFloat(self):
+    """Reads and returns a 4-byte floating-point number."""
+    serialized = self._stream.ReadBytes(4)
+    return struct.unpack(wire_format.FORMAT_FLOAT_LITTLE_ENDIAN, serialized)[0]
+
+  def ReadDouble(self):
+    """Reads and returns an 8-byte floating-point number."""
+    serialized = self._stream.ReadBytes(8)
+    return struct.unpack(wire_format.FORMAT_DOUBLE_LITTLE_ENDIAN, serialized)[0]
+
+  def ReadBool(self):
+    """Reads and returns a bool."""
+    i = self._stream.ReadVarUInt32()
+    return bool(i)
+
+  def ReadEnum(self):
+    """Reads and returns an enum value."""
+    return self._stream.ReadVarUInt32()
+
+  def ReadString(self):
+    """Reads and returns a length-delimited string."""
+    bytes = self.ReadBytes()
+    return unicode(bytes, 'utf-8')
+
+  def ReadBytes(self):
+    """Reads and returns a length-delimited byte sequence."""
+    length = self._stream.ReadVarUInt32()
+    return self._stream.ReadBytes(length)
+
+  def ReadMessageInto(self, msg):
+    """Calls msg.MergeFromString() to merge
+    length-delimited serialized message data into |msg|.
+
+    REQUIRES: The decoder must be positioned at the serialized "length"
+      prefix to a length-delmiited serialized message.
+
+    POSTCONDITION: The decoder is positioned just after the
+      serialized message, and we have merged those serialized
+      contents into |msg|.
+    """
+    length = self._stream.ReadVarUInt32()
+    sub_buffer = self._stream.GetSubBuffer(length)
+    num_bytes_used = msg.MergeFromString(sub_buffer)
+    if num_bytes_used != length:
+      raise message.DecodeError(
+          'Submessage told to deserialize from %d-byte encoding, '
+          'but used only %d bytes' % (length, num_bytes_used))
+    self._stream.SkipBytes(num_bytes_used)
+
+  def ReadGroupInto(self, expected_field_number, group):
+    """Calls group.MergeFromString() to merge
+    END_GROUP-delimited serialized message data into |group|.
+    We'll raise an exception if we don't find an END_GROUP
+    tag immediately after the serialized message contents.
+
+    REQUIRES: The decoder is positioned just after the START_GROUP
+      tag for this group.
+
+    POSTCONDITION: The decoder is positioned just after the
+      END_GROUP tag for this group, and we have merged
+      the contents of the group into |group|.
+    """
+    sub_buffer = self._stream.GetSubBuffer()  # No a priori length limit.
+    num_bytes_used = group.MergeFromString(sub_buffer)
+    if num_bytes_used < 0:
+      raise message.DecodeError('Group message reported negative bytes read.')
+    self._stream.SkipBytes(num_bytes_used)
+    field_number, field_type = self.ReadFieldNumberAndWireType()
+    if field_type != wire_format.WIRETYPE_END_GROUP:
+      raise message.DecodeError('Group message did not end with an END_GROUP.')
+    if field_number != expected_field_number:
+      raise message.DecodeError('END_GROUP tag had field '
+                                'number %d, was expecting field number %d' % (
+          field_number, expected_field_number))
+    # We're now positioned just after the END_GROUP tag.  Perfect.
diff --git a/python/google/protobuf/internal/decoder_test.py b/python/google/protobuf/internal/decoder_test.py
new file mode 100755
index 0000000..98e4647
--- /dev/null
+++ b/python/google/protobuf/internal/decoder_test.py
@@ -0,0 +1,256 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.internal.decoder."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import struct
+import unittest
+from google.protobuf.internal import decoder
+from google.protobuf.internal import encoder
+from google.protobuf.internal import input_stream
+from google.protobuf.internal import wire_format
+from google.protobuf import message
+import logging
+import mox
+
+
+class DecoderTest(unittest.TestCase):
+
+  def setUp(self):
+    self.mox = mox.Mox()
+    self.mock_stream = self.mox.CreateMock(input_stream.InputStream)
+    self.mock_message = self.mox.CreateMock(message.Message)
+
+  def testReadFieldNumberAndWireType(self):
+    # Test field numbers that will require various varint sizes.
+    for expected_field_number in (1, 15, 16, 2047, 2048):
+      for expected_wire_type in range(6):  # Highest-numbered wiretype is 5.
+        e = encoder.Encoder()
+        e.AppendTag(expected_field_number, expected_wire_type)
+        s = e.ToString()
+        d = decoder.Decoder(s)
+        field_number, wire_type = d.ReadFieldNumberAndWireType()
+        self.assertEqual(expected_field_number, field_number)
+        self.assertEqual(expected_wire_type, wire_type)
+
+  def ReadScalarTestHelper(self, test_name, decoder_method, expected_result,
+                           expected_stream_method_name,
+                           stream_method_return, *args):
+    """Helper for testReadScalars below.
+
+    Calls one of the Decoder.Read*() methods and ensures that the results are
+    as expected.
+
+    Args:
+      test_name: Name of this test, used for logging only.
+      decoder_method: Unbound decoder.Decoder method to call.
+      expected_result: Value we expect returned from decoder_method().
+      expected_stream_method_name: (string) Name of the InputStream
+        method we expect Decoder to call to actually read the value
+        on the wire.
+      stream_method_return: Value our mocked-out stream method should
+        return to the decoder.
+      args: Additional arguments that we expect to be passed to the
+        stream method.
+    """
+    logging.info('Testing %s scalar input.\n'
+                 'Calling %r(), and expecting that to call the '
+                 'stream method %s(%r), which will return %r.  Finally, '
+                 'expecting the Decoder method to return %r'% (
+        test_name, decoder_method,
+        expected_stream_method_name, args, stream_method_return,
+        expected_result))
+
+    d = decoder.Decoder('')
+    d._stream = self.mock_stream
+    if decoder_method in (decoder.Decoder.ReadString,
+                          decoder.Decoder.ReadBytes):
+      self.mock_stream.ReadVarUInt32().AndReturn(len(stream_method_return))
+    # We have to use names instead of methods to work around some
+    # mox weirdness.  (ResetAll() is overzealous).
+    expected_stream_method = getattr(self.mock_stream,
+                                     expected_stream_method_name)
+    expected_stream_method(*args).AndReturn(stream_method_return)
+
+    self.mox.ReplayAll()
+    result = decoder_method(d)
+    self.assertEqual(expected_result, result)
+    self.assert_(isinstance(result, type(expected_result)))
+    self.mox.VerifyAll()
+    self.mox.ResetAll()
+
+  VAL = 1.125  # Perfectly representable as a float (no rounding error).
+  LITTLE_FLOAT_VAL = '\x00\x00\x90?'
+  LITTLE_DOUBLE_VAL = '\x00\x00\x00\x00\x00\x00\xf2?'
+
+  def testReadScalars(self):
+    test_string = 'I can feel myself getting sutpider.'
+    scalar_tests = [
+        ['int32', decoder.Decoder.ReadInt32, 0, 'ReadVarint32', 0],
+        ['int64', decoder.Decoder.ReadInt64, 0, 'ReadVarint64', 0],
+        ['uint32', decoder.Decoder.ReadUInt32, 0, 'ReadVarUInt32', 0],
+        ['uint64', decoder.Decoder.ReadUInt64, 0, 'ReadVarUInt64', 0],
+        ['fixed32', decoder.Decoder.ReadFixed32, 0xffffffff,
+         'ReadLittleEndian32', 0xffffffff],
+        ['fixed64', decoder.Decoder.ReadFixed64, 0xffffffffffffffff,
+        'ReadLittleEndian64', 0xffffffffffffffff],
+        ['sfixed32', decoder.Decoder.ReadSFixed32, long(-1),
+         'ReadLittleEndian32', long(0xffffffff)],
+        ['sfixed64', decoder.Decoder.ReadSFixed64, long(-1),
+         'ReadLittleEndian64', 0xffffffffffffffff],
+        ['float', decoder.Decoder.ReadFloat, self.VAL,
+         'ReadBytes', self.LITTLE_FLOAT_VAL, 4],
+        ['double', decoder.Decoder.ReadDouble, self.VAL,
+         'ReadBytes', self.LITTLE_DOUBLE_VAL, 8],
+        ['bool', decoder.Decoder.ReadBool, True, 'ReadVarUInt32', 1],
+        ['enum', decoder.Decoder.ReadEnum, 23, 'ReadVarUInt32', 23],
+        ['string', decoder.Decoder.ReadString,
+         unicode(test_string, 'utf-8'), 'ReadBytes', test_string,
+         len(test_string)],
+        ['utf8-string', decoder.Decoder.ReadString,
+         unicode(test_string, 'utf-8'), 'ReadBytes', test_string,
+         len(test_string)],
+        ['bytes', decoder.Decoder.ReadBytes,
+         test_string, 'ReadBytes', test_string, len(test_string)],
+        # We test zigzag decoding routines more extensively below.
+        ['sint32', decoder.Decoder.ReadSInt32, -1, 'ReadVarUInt32', 1],
+        ['sint64', decoder.Decoder.ReadSInt64, -1, 'ReadVarUInt64', 1],
+        ]
+    # Ensure that we're testing different Decoder methods and using
+    # different test names in all test cases above.
+    self.assertEqual(len(scalar_tests), len(set(t[0] for t in scalar_tests)))
+    self.assert_(len(scalar_tests) >= len(set(t[1] for t in scalar_tests)))
+    for args in scalar_tests:
+      self.ReadScalarTestHelper(*args)
+
+  def testReadMessageInto(self):
+    length = 23
+    def Test(simulate_error):
+      d = decoder.Decoder('')
+      d._stream = self.mock_stream
+      self.mock_stream.ReadVarUInt32().AndReturn(length)
+      sub_buffer = object()
+      self.mock_stream.GetSubBuffer(length).AndReturn(sub_buffer)
+
+      if simulate_error:
+        self.mock_message.MergeFromString(sub_buffer).AndReturn(length - 1)
+        self.mox.ReplayAll()
+        self.assertRaises(
+            message.DecodeError, d.ReadMessageInto, self.mock_message)
+      else:
+        self.mock_message.MergeFromString(sub_buffer).AndReturn(length)
+        self.mock_stream.SkipBytes(length)
+        self.mox.ReplayAll()
+        d.ReadMessageInto(self.mock_message)
+
+      self.mox.VerifyAll()
+      self.mox.ResetAll()
+
+    Test(simulate_error=False)
+    Test(simulate_error=True)
+
+  def testReadGroupInto_Success(self):
+    # Test both the empty and nonempty cases.
+    for num_bytes in (5, 0):
+      field_number = expected_field_number = 10
+      d = decoder.Decoder('')
+      d._stream = self.mock_stream
+      sub_buffer = object()
+      self.mock_stream.GetSubBuffer().AndReturn(sub_buffer)
+      self.mock_message.MergeFromString(sub_buffer).AndReturn(num_bytes)
+      self.mock_stream.SkipBytes(num_bytes)
+      self.mock_stream.ReadVarUInt32().AndReturn(wire_format.PackTag(
+          field_number, wire_format.WIRETYPE_END_GROUP))
+      self.mox.ReplayAll()
+      d.ReadGroupInto(expected_field_number, self.mock_message)
+      self.mox.VerifyAll()
+      self.mox.ResetAll()
+
+  def ReadGroupInto_FailureTestHelper(self, bytes_read):
+    d = decoder.Decoder('')
+    d._stream = self.mock_stream
+    sub_buffer = object()
+    self.mock_stream.GetSubBuffer().AndReturn(sub_buffer)
+    self.mock_message.MergeFromString(sub_buffer).AndReturn(bytes_read)
+    return d
+
+  def testReadGroupInto_NegativeBytesReported(self):
+    expected_field_number = 10
+    d = self.ReadGroupInto_FailureTestHelper(bytes_read=-1)
+    self.mox.ReplayAll()
+    self.assertRaises(message.DecodeError,
+                      d.ReadGroupInto, expected_field_number,
+                      self.mock_message)
+    self.mox.VerifyAll()
+
+  def testReadGroupInto_NoEndGroupTag(self):
+    field_number = expected_field_number = 10
+    num_bytes = 5
+    d = self.ReadGroupInto_FailureTestHelper(bytes_read=num_bytes)
+    self.mock_stream.SkipBytes(num_bytes)
+    # Right field number, wrong wire type.
+    self.mock_stream.ReadVarUInt32().AndReturn(wire_format.PackTag(
+        field_number, wire_format.WIRETYPE_LENGTH_DELIMITED))
+    self.mox.ReplayAll()
+    self.assertRaises(message.DecodeError,
+                      d.ReadGroupInto, expected_field_number,
+                      self.mock_message)
+    self.mox.VerifyAll()
+
+  def testReadGroupInto_WrongFieldNumberInEndGroupTag(self):
+    expected_field_number = 10
+    field_number = expected_field_number + 1
+    num_bytes = 5
+    d = self.ReadGroupInto_FailureTestHelper(bytes_read=num_bytes)
+    self.mock_stream.SkipBytes(num_bytes)
+    # Wrong field number, right wire type.
+    self.mock_stream.ReadVarUInt32().AndReturn(wire_format.PackTag(
+        field_number, wire_format.WIRETYPE_END_GROUP))
+    self.mox.ReplayAll()
+    self.assertRaises(message.DecodeError,
+                      d.ReadGroupInto, expected_field_number,
+                      self.mock_message)
+    self.mox.VerifyAll()
+
+  def testSkipBytes(self):
+    d = decoder.Decoder('')
+    num_bytes = 1024
+    self.mock_stream.SkipBytes(num_bytes)
+    d._stream = self.mock_stream
+    self.mox.ReplayAll()
+    d.SkipBytes(num_bytes)
+    self.mox.VerifyAll()
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/descriptor_test.py b/python/google/protobuf/internal/descriptor_test.py
new file mode 100755
index 0000000..eb9f2be
--- /dev/null
+++ b/python/google/protobuf/internal/descriptor_test.py
@@ -0,0 +1,113 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Unittest for google.protobuf.internal.descriptor."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import unittest
+from google.protobuf import descriptor_pb2
+from google.protobuf import descriptor
+
+class DescriptorTest(unittest.TestCase):
+
+  def setUp(self):
+    self.my_enum = descriptor.EnumDescriptor(
+        name='ForeignEnum',
+        full_name='protobuf_unittest.ForeignEnum',
+        filename='ForeignEnum',
+        values=[
+          descriptor.EnumValueDescriptor(name='FOREIGN_FOO', index=0, number=4),
+          descriptor.EnumValueDescriptor(name='FOREIGN_BAR', index=1, number=5),
+          descriptor.EnumValueDescriptor(name='FOREIGN_BAZ', index=2, number=6),
+        ])
+    self.my_message = descriptor.Descriptor(
+        name='NestedMessage',
+        full_name='protobuf_unittest.TestAllTypes.NestedMessage',
+        filename='some/filename/some.proto',
+        containing_type=None,
+        fields=[
+          descriptor.FieldDescriptor(
+            name='bb',
+            full_name='protobuf_unittest.TestAllTypes.NestedMessage.bb',
+            index=0, number=1,
+            type=5, cpp_type=1, label=1,
+            default_value=0,
+            message_type=None, enum_type=None, containing_type=None,
+            is_extension=False, extension_scope=None),
+        ],
+        nested_types=[],
+        enum_types=[
+          self.my_enum,
+        ],
+        extensions=[])
+    self.my_method = descriptor.MethodDescriptor(
+        name='Bar',
+        full_name='protobuf_unittest.TestService.Bar',
+        index=0,
+        containing_service=None,
+        input_type=None,
+        output_type=None)
+    self.my_service = descriptor.ServiceDescriptor(
+        name='TestServiceWithOptions',
+        full_name='protobuf_unittest.TestServiceWithOptions',
+        index=0,
+        methods=[
+            self.my_method
+        ])
+
+  def testEnumFixups(self):
+    self.assertEqual(self.my_enum, self.my_enum.values[0].type)
+
+  def testContainingTypeFixups(self):
+    self.assertEqual(self.my_message, self.my_message.fields[0].containing_type)
+    self.assertEqual(self.my_message, self.my_enum.containing_type)
+
+  def testContainingServiceFixups(self):
+    self.assertEqual(self.my_service, self.my_method.containing_service)
+
+  def testGetOptions(self):
+    self.assertEqual(self.my_enum.GetOptions(),
+                     descriptor_pb2.EnumOptions())
+    self.assertEqual(self.my_enum.values[0].GetOptions(),
+                     descriptor_pb2.EnumValueOptions())
+    self.assertEqual(self.my_message.GetOptions(),
+                     descriptor_pb2.MessageOptions())
+    self.assertEqual(self.my_message.fields[0].GetOptions(),
+                     descriptor_pb2.FieldOptions())
+    self.assertEqual(self.my_method.GetOptions(),
+                     descriptor_pb2.MethodOptions())
+    self.assertEqual(self.my_service.GetOptions(),
+                     descriptor_pb2.ServiceOptions())
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/encoder.py b/python/google/protobuf/internal/encoder.py
new file mode 100755
index 0000000..3ec3b2b
--- /dev/null
+++ b/python/google/protobuf/internal/encoder.py
@@ -0,0 +1,280 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Class for encoding protocol message primitives.
+
+Contains the logic for encoding every logical protocol field type
+into one of the 5 physical wire types.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import struct
+from google.protobuf import message
+from google.protobuf.internal import wire_format
+from google.protobuf.internal import output_stream
+
+
+# Note that much of this code is ported from //net/proto/ProtocolBuffer, and
+# that the interface is strongly inspired by WireFormat from the C++ proto2
+# implementation.
+
+
+class Encoder(object):
+
+  """Encodes logical protocol buffer fields to the wire format."""
+
+  def __init__(self):
+    self._stream = output_stream.OutputStream()
+
+  def ToString(self):
+    """Returns all values encoded in this object as a string."""
+    return self._stream.ToString()
+
+  # Append*NoTag methods.  These are necessary for serializing packed
+  # repeated fields.  The Append*() methods call these methods to do
+  # the actual serialization.
+  def AppendInt32NoTag(self, value):
+    """Appends a 32-bit integer to our buffer, varint-encoded."""
+    self._stream.AppendVarint32(value)
+
+  def AppendInt64NoTag(self, value):
+    """Appends a 64-bit integer to our buffer, varint-encoded."""
+    self._stream.AppendVarint64(value)
+
+  def AppendUInt32NoTag(self, unsigned_value):
+    """Appends an unsigned 32-bit integer to our buffer, varint-encoded."""
+    self._stream.AppendVarUInt32(unsigned_value)
+
+  def AppendUInt64NoTag(self, unsigned_value):
+    """Appends an unsigned 64-bit integer to our buffer, varint-encoded."""
+    self._stream.AppendVarUInt64(unsigned_value)
+
+  def AppendSInt32NoTag(self, value):
+    """Appends a 32-bit integer to our buffer, zigzag-encoded and then
+    varint-encoded.
+    """
+    zigzag_value = wire_format.ZigZagEncode(value)
+    self._stream.AppendVarUInt32(zigzag_value)
+
+  def AppendSInt64NoTag(self, value):
+    """Appends a 64-bit integer to our buffer, zigzag-encoded and then
+    varint-encoded.
+    """
+    zigzag_value = wire_format.ZigZagEncode(value)
+    self._stream.AppendVarUInt64(zigzag_value)
+
+  def AppendFixed32NoTag(self, unsigned_value):
+    """Appends an unsigned 32-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self._stream.AppendLittleEndian32(unsigned_value)
+
+  def AppendFixed64NoTag(self, unsigned_value):
+    """Appends an unsigned 64-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self._stream.AppendLittleEndian64(unsigned_value)
+
+  def AppendSFixed32NoTag(self, value):
+    """Appends a signed 32-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    sign = (value & 0x80000000) and -1 or 0
+    if value >> 32 != sign:
+      raise message.EncodeError('SFixed32 out of range: %d' % value)
+    self._stream.AppendLittleEndian32(value & 0xffffffff)
+
+  def AppendSFixed64NoTag(self, value):
+    """Appends a signed 64-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    sign = (value & 0x8000000000000000) and -1 or 0
+    if value >> 64 != sign:
+      raise message.EncodeError('SFixed64 out of range: %d' % value)
+    self._stream.AppendLittleEndian64(value & 0xffffffffffffffff)
+
+  def AppendFloatNoTag(self, value):
+    """Appends a floating-point number to our buffer."""
+    self._stream.AppendRawBytes(
+        struct.pack(wire_format.FORMAT_FLOAT_LITTLE_ENDIAN, value))
+
+  def AppendDoubleNoTag(self, value):
+    """Appends a double-precision floating-point number to our buffer."""
+    self._stream.AppendRawBytes(
+        struct.pack(wire_format.FORMAT_DOUBLE_LITTLE_ENDIAN, value))
+
+  def AppendBoolNoTag(self, value):
+    """Appends a boolean to our buffer."""
+    self.AppendInt32NoTag(value)
+
+  def AppendEnumNoTag(self, value):
+    """Appends an enum value to our buffer."""
+    self.AppendInt32NoTag(value)
+
+
+  # All the Append*() methods below first append a tag+type pair to the buffer
+  # before appending the specified value.
+
+  def AppendInt32(self, field_number, value):
+    """Appends a 32-bit integer to our buffer, varint-encoded."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendInt32NoTag(value)
+
+  def AppendInt64(self, field_number, value):
+    """Appends a 64-bit integer to our buffer, varint-encoded."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendInt64NoTag(value)
+
+  def AppendUInt32(self, field_number, unsigned_value):
+    """Appends an unsigned 32-bit integer to our buffer, varint-encoded."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendUInt32NoTag(unsigned_value)
+
+  def AppendUInt64(self, field_number, unsigned_value):
+    """Appends an unsigned 64-bit integer to our buffer, varint-encoded."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendUInt64NoTag(unsigned_value)
+
+  def AppendSInt32(self, field_number, value):
+    """Appends a 32-bit integer to our buffer, zigzag-encoded and then
+    varint-encoded.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendSInt32NoTag(value)
+
+  def AppendSInt64(self, field_number, value):
+    """Appends a 64-bit integer to our buffer, zigzag-encoded and then
+    varint-encoded.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_VARINT)
+    self.AppendSInt64NoTag(value)
+
+  def AppendFixed32(self, field_number, unsigned_value):
+    """Appends an unsigned 32-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED32)
+    self.AppendFixed32NoTag(unsigned_value)
+
+  def AppendFixed64(self, field_number, unsigned_value):
+    """Appends an unsigned 64-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED64)
+    self.AppendFixed64NoTag(unsigned_value)
+
+  def AppendSFixed32(self, field_number, value):
+    """Appends a signed 32-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED32)
+    self.AppendSFixed32NoTag(value)
+
+  def AppendSFixed64(self, field_number, value):
+    """Appends a signed 64-bit integer to our buffer, in little-endian
+    byte-order.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED64)
+    self.AppendSFixed64NoTag(value)
+
+  def AppendFloat(self, field_number, value):
+    """Appends a floating-point number to our buffer."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED32)
+    self.AppendFloatNoTag(value)
+
+  def AppendDouble(self, field_number, value):
+    """Appends a double-precision floating-point number to our buffer."""
+    self.AppendTag(field_number, wire_format.WIRETYPE_FIXED64)
+    self.AppendDoubleNoTag(value)
+
+  def AppendBool(self, field_number, value):
+    """Appends a boolean to our buffer."""
+    self.AppendInt32(field_number, value)
+
+  def AppendEnum(self, field_number, value):
+    """Appends an enum value to our buffer."""
+    self.AppendInt32(field_number, value)
+
+  def AppendString(self, field_number, value):
+    """Appends a length-prefixed unicode string, encoded as UTF-8 to our buffer,
+    with the length varint-encoded.
+    """
+    self.AppendBytes(field_number, value.encode('utf-8'))
+
+  def AppendBytes(self, field_number, value):
+    """Appends a length-prefixed sequence of bytes to our buffer, with the
+    length varint-encoded.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_LENGTH_DELIMITED)
+    self._stream.AppendVarUInt32(len(value))
+    self._stream.AppendRawBytes(value)
+
+  # TODO(robinson): For AppendGroup() and AppendMessage(), we'd really like to
+  # avoid the extra string copy here.  We can do so if we widen the Message
+  # interface to be able to serialize to a stream in addition to a string.  The
+  # challenge when thinking ahead to the Python/C API implementation of Message
+  # is finding a stream-like Python thing to which we can write raw bytes
+  # from C.  I'm not sure such a thing exists(?).  (array.array is pretty much
+  # what we want, but it's not directly exposed in the Python/C API).
+
+  def AppendGroup(self, field_number, group):
+    """Appends a group to our buffer.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_START_GROUP)
+    self._stream.AppendRawBytes(group.SerializeToString())
+    self.AppendTag(field_number, wire_format.WIRETYPE_END_GROUP)
+
+  def AppendMessage(self, field_number, msg):
+    """Appends a nested message to our buffer.
+    """
+    self.AppendTag(field_number, wire_format.WIRETYPE_LENGTH_DELIMITED)
+    self._stream.AppendVarUInt32(msg.ByteSize())
+    self._stream.AppendRawBytes(msg.SerializeToString())
+
+  def AppendMessageSetItem(self, field_number, msg):
+    """Appends an item using the message set wire format.
+
+    The message set message looks like this:
+      message MessageSet {
+        repeated group Item = 1 {
+          required int32 type_id = 2;
+          required string message = 3;
+        }
+      }
+    """
+    self.AppendTag(1, wire_format.WIRETYPE_START_GROUP)
+    self.AppendInt32(2, field_number)
+    self.AppendMessage(3, msg)
+    self.AppendTag(1, wire_format.WIRETYPE_END_GROUP)
+
+  def AppendTag(self, field_number, wire_type):
+    """Appends a tag containing field number and wire type information."""
+    self._stream.AppendVarUInt32(wire_format.PackTag(field_number, wire_type))
diff --git a/python/google/protobuf/internal/encoder_test.py b/python/google/protobuf/internal/encoder_test.py
new file mode 100755
index 0000000..bf75ea8
--- /dev/null
+++ b/python/google/protobuf/internal/encoder_test.py
@@ -0,0 +1,286 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.internal.encoder."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import struct
+import logging
+import unittest
+from google.protobuf.internal import wire_format
+from google.protobuf.internal import encoder
+from google.protobuf.internal import output_stream
+from google.protobuf import message
+import mox
+
+
+class EncoderTest(unittest.TestCase):
+
+  def setUp(self):
+    self.mox = mox.Mox()
+    self.encoder = encoder.Encoder()
+    self.mock_stream = self.mox.CreateMock(output_stream.OutputStream)
+    self.mock_message = self.mox.CreateMock(message.Message)
+    self.encoder._stream = self.mock_stream
+
+  def PackTag(self, field_number, wire_type):
+    return wire_format.PackTag(field_number, wire_type)
+
+  def AppendScalarTestHelper(self, test_name, encoder_method,
+                             expected_stream_method_name,
+                             wire_type, field_value,
+                             expected_value=None, expected_length=None,
+                             is_tag_test=True):
+    """Helper for testAppendScalars.
+
+    Calls one of the Encoder methods, and ensures that the Encoder
+    in turn makes the expected calls into its OutputStream.
+
+    Args:
+      test_name: Name of this test, used only for logging.
+      encoder_method: Callable on self.encoder. This is the Encoder
+        method we're testing.  If is_tag_test=True, the encoder method
+        accepts a field_number and field_value. if is_tag_test=False,
+        the encoder method accepts a field_value.
+      expected_stream_method_name: (string) Name of the OutputStream
+        method we expect Encoder to call to actually put the value
+        on the wire.
+      wire_type: The WIRETYPE_* constant we expect encoder to
+        use in the specified encoder_method.
+      field_value: The value we're trying to encode.  Passed
+        into encoder_method.
+      expected_value: The value we expect Encoder to pass into
+        the OutputStream method.  If None, we expect field_value
+        to pass through unmodified.
+      expected_length: The length we expect Encoder to pass to the
+        AppendVarUInt32 method. If None we expect the length of the
+        field_value.
+      is_tag_test: A Boolean.  If True (the default), we append the
+        the packed field number and wire_type to the stream before
+        the field value.
+    """
+    if expected_value is None:
+      expected_value = field_value
+
+    logging.info('Testing %s scalar output.\n'
+                 'Calling %r(%r), and expecting that to call the '
+                 'stream method %s(%r).' % (
+        test_name, encoder_method, field_value,
+        expected_stream_method_name, expected_value))
+
+    if is_tag_test:
+      field_number = 10
+      # Should first append the field number and type information.
+      self.mock_stream.AppendVarUInt32(self.PackTag(field_number, wire_type))
+      # If we're length-delimited, we should then append the length.
+      if wire_type == wire_format.WIRETYPE_LENGTH_DELIMITED:
+        if expected_length is None:
+          expected_length = len(field_value)
+        self.mock_stream.AppendVarUInt32(expected_length)
+
+    # Should then append the value itself.
+    # We have to use names instead of methods to work around some
+    # mox weirdness.  (ResetAll() is overzealous).
+    expected_stream_method = getattr(self.mock_stream,
+                                     expected_stream_method_name)
+    expected_stream_method(expected_value)
+
+    self.mox.ReplayAll()
+    if is_tag_test:
+      encoder_method(field_number, field_value)
+    else:
+      encoder_method(field_value)
+    self.mox.VerifyAll()
+    self.mox.ResetAll()
+
+  VAL = 1.125  # Perfectly representable as a float (no rounding error).
+  LITTLE_FLOAT_VAL = '\x00\x00\x90?'
+  LITTLE_DOUBLE_VAL = '\x00\x00\x00\x00\x00\x00\xf2?'
+
+  def testAppendScalars(self):
+    utf8_bytes = '\xd0\xa2\xd0\xb5\xd1\x81\xd1\x82'
+    utf8_string = unicode(utf8_bytes, 'utf-8')
+    scalar_tests = [
+        ['int32', self.encoder.AppendInt32, 'AppendVarint32',
+         wire_format.WIRETYPE_VARINT, 0],
+        ['int64', self.encoder.AppendInt64, 'AppendVarint64',
+         wire_format.WIRETYPE_VARINT, 0],
+        ['uint32', self.encoder.AppendUInt32, 'AppendVarUInt32',
+         wire_format.WIRETYPE_VARINT, 0],
+        ['uint64', self.encoder.AppendUInt64, 'AppendVarUInt64',
+         wire_format.WIRETYPE_VARINT, 0],
+        ['fixed32', self.encoder.AppendFixed32, 'AppendLittleEndian32',
+         wire_format.WIRETYPE_FIXED32, 0],
+        ['fixed64', self.encoder.AppendFixed64, 'AppendLittleEndian64',
+         wire_format.WIRETYPE_FIXED64, 0],
+        ['sfixed32', self.encoder.AppendSFixed32, 'AppendLittleEndian32',
+         wire_format.WIRETYPE_FIXED32, -1, 0xffffffff],
+        ['sfixed64', self.encoder.AppendSFixed64, 'AppendLittleEndian64',
+         wire_format.WIRETYPE_FIXED64, -1, 0xffffffffffffffff],
+        ['float', self.encoder.AppendFloat, 'AppendRawBytes',
+         wire_format.WIRETYPE_FIXED32, self.VAL, self.LITTLE_FLOAT_VAL],
+        ['double', self.encoder.AppendDouble, 'AppendRawBytes',
+         wire_format.WIRETYPE_FIXED64, self.VAL, self.LITTLE_DOUBLE_VAL],
+        ['bool', self.encoder.AppendBool, 'AppendVarint32',
+         wire_format.WIRETYPE_VARINT, False],
+        ['enum', self.encoder.AppendEnum, 'AppendVarint32',
+         wire_format.WIRETYPE_VARINT, 0],
+        ['string', self.encoder.AppendString, 'AppendRawBytes',
+         wire_format.WIRETYPE_LENGTH_DELIMITED,
+         "You're in a maze of twisty little passages, all alike."],
+        ['utf8-string', self.encoder.AppendString, 'AppendRawBytes',
+         wire_format.WIRETYPE_LENGTH_DELIMITED, utf8_string,
+         utf8_bytes, len(utf8_bytes)],
+        # We test zigzag encoding routines more extensively below.
+        ['sint32', self.encoder.AppendSInt32, 'AppendVarUInt32',
+         wire_format.WIRETYPE_VARINT, -1, 1],
+        ['sint64', self.encoder.AppendSInt64, 'AppendVarUInt64',
+         wire_format.WIRETYPE_VARINT, -1, 1],
+        ]
+    # Ensure that we're testing different Encoder methods and using
+    # different test names in all test cases above.
+    self.assertEqual(len(scalar_tests), len(set(t[0] for t in scalar_tests)))
+    self.assert_(len(scalar_tests) >= len(set(t[1] for t in scalar_tests)))
+    for args in scalar_tests:
+      self.AppendScalarTestHelper(*args)
+
+  def testAppendScalarsWithoutTags(self):
+    scalar_no_tag_tests = [
+        ['int32', self.encoder.AppendInt32NoTag, 'AppendVarint32', None, 0],
+        ['int64', self.encoder.AppendInt64NoTag, 'AppendVarint64', None, 0],
+        ['uint32', self.encoder.AppendUInt32NoTag, 'AppendVarUInt32', None, 0],
+        ['uint64', self.encoder.AppendUInt64NoTag, 'AppendVarUInt64', None, 0],
+        ['fixed32', self.encoder.AppendFixed32NoTag,
+         'AppendLittleEndian32', None, 0],
+        ['fixed64', self.encoder.AppendFixed64NoTag,
+         'AppendLittleEndian64', None, 0],
+        ['sfixed32', self.encoder.AppendSFixed32NoTag,
+         'AppendLittleEndian32', None, 0],
+        ['sfixed64', self.encoder.AppendSFixed64NoTag,
+         'AppendLittleEndian64', None, 0],
+        ['float', self.encoder.AppendFloatNoTag,
+         'AppendRawBytes', None, self.VAL, self.LITTLE_FLOAT_VAL],
+        ['double', self.encoder.AppendDoubleNoTag,
+         'AppendRawBytes', None, self.VAL, self.LITTLE_DOUBLE_VAL],
+        ['bool', self.encoder.AppendBoolNoTag, 'AppendVarint32', None, 0],
+        ['enum', self.encoder.AppendEnumNoTag, 'AppendVarint32', None, 0],
+        ['sint32', self.encoder.AppendSInt32NoTag,
+         'AppendVarUInt32', None, -1, 1],
+        ['sint64', self.encoder.AppendSInt64NoTag,
+         'AppendVarUInt64', None, -1, 1],
+    ]
+
+    self.assertEqual(len(scalar_no_tag_tests),
+                     len(set(t[0] for t in scalar_no_tag_tests)))
+    self.assert_(len(scalar_no_tag_tests) >=
+                 len(set(t[1] for t in scalar_no_tag_tests)))
+    for args in scalar_no_tag_tests:
+      # For no tag tests, the wire_type is not used, so we put in None.
+      self.AppendScalarTestHelper(is_tag_test=False, *args)
+
+  def testAppendGroup(self):
+    field_number = 23
+    # Should first append the start-group marker.
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(field_number, wire_format.WIRETYPE_START_GROUP))
+    # Should then serialize itself.
+    self.mock_message.SerializeToString().AndReturn('foo')
+    self.mock_stream.AppendRawBytes('foo')
+    # Should finally append the end-group marker.
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(field_number, wire_format.WIRETYPE_END_GROUP))
+
+    self.mox.ReplayAll()
+    self.encoder.AppendGroup(field_number, self.mock_message)
+    self.mox.VerifyAll()
+
+  def testAppendMessage(self):
+    field_number = 23
+    byte_size = 42
+    # Should first append the field number and type information.
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(field_number, wire_format.WIRETYPE_LENGTH_DELIMITED))
+    # Should then append its length.
+    self.mock_message.ByteSize().AndReturn(byte_size)
+    self.mock_stream.AppendVarUInt32(byte_size)
+    # Should then serialize itself to the encoder.
+    self.mock_message.SerializeToString().AndReturn('foo')
+    self.mock_stream.AppendRawBytes('foo')
+
+    self.mox.ReplayAll()
+    self.encoder.AppendMessage(field_number, self.mock_message)
+    self.mox.VerifyAll()
+
+  def testAppendMessageSetItem(self):
+    field_number = 23
+    byte_size = 42
+    # Should first append the field number and type information.
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(1, wire_format.WIRETYPE_START_GROUP))
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(2, wire_format.WIRETYPE_VARINT))
+    self.mock_stream.AppendVarint32(field_number)
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(3, wire_format.WIRETYPE_LENGTH_DELIMITED))
+    # Should then append its length.
+    self.mock_message.ByteSize().AndReturn(byte_size)
+    self.mock_stream.AppendVarUInt32(byte_size)
+    # Should then serialize itself to the encoder.
+    self.mock_message.SerializeToString().AndReturn('foo')
+    self.mock_stream.AppendRawBytes('foo')
+    self.mock_stream.AppendVarUInt32(
+        self.PackTag(1, wire_format.WIRETYPE_END_GROUP))
+
+    self.mox.ReplayAll()
+    self.encoder.AppendMessageSetItem(field_number, self.mock_message)
+    self.mox.VerifyAll()
+
+  def testAppendSFixed(self):
+    # Most of our bounds-checking is done in output_stream.py,
+    # but encoder.py is responsible for transforming signed
+    # fixed-width integers into unsigned ones, so we test here
+    # to ensure that we're not losing any entropy when we do
+    # that conversion.
+    field_number = 10
+    self.assertRaises(message.EncodeError, self.encoder.AppendSFixed32,
+        10, wire_format.UINT32_MAX + 1)
+    self.assertRaises(message.EncodeError, self.encoder.AppendSFixed32,
+        10, -(1 << 32))
+    self.assertRaises(message.EncodeError, self.encoder.AppendSFixed64,
+        10, wire_format.UINT64_MAX + 1)
+    self.assertRaises(message.EncodeError, self.encoder.AppendSFixed64,
+        10, -(1 << 64))
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/generator_test.py b/python/google/protobuf/internal/generator_test.py
new file mode 100755
index 0000000..11fcfa0
--- /dev/null
+++ b/python/google/protobuf/internal/generator_test.py
@@ -0,0 +1,100 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# TODO(robinson): Flesh this out considerably.  We focused on reflection_test.py
+# first, since it's testing the subtler code, and since it provides decent
+# indirect testing of the protocol compiler output.
+
+"""Unittest that directly tests the output of the pure-Python protocol
+compiler.  See //net/proto2/internal/reflection_test.py for a test which
+further ensures that we can use Python protocol message objects as we expect.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import unittest
+from google.protobuf import unittest_mset_pb2
+from google.protobuf import unittest_pb2
+
+
+class GeneratorTest(unittest.TestCase):
+
+  def testNestedMessageDescriptor(self):
+    field_name = 'optional_nested_message'
+    proto_type = unittest_pb2.TestAllTypes
+    self.assertEqual(
+        proto_type.NestedMessage.DESCRIPTOR,
+        proto_type.DESCRIPTOR.fields_by_name[field_name].message_type)
+
+  def testEnums(self):
+    # We test only module-level enums here.
+    # TODO(robinson): Examine descriptors directly to check
+    # enum descriptor output.
+    self.assertEqual(4, unittest_pb2.FOREIGN_FOO)
+    self.assertEqual(5, unittest_pb2.FOREIGN_BAR)
+    self.assertEqual(6, unittest_pb2.FOREIGN_BAZ)
+
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(1, proto.FOO)
+    self.assertEqual(1, unittest_pb2.TestAllTypes.FOO)
+    self.assertEqual(2, proto.BAR)
+    self.assertEqual(2, unittest_pb2.TestAllTypes.BAR)
+    self.assertEqual(3, proto.BAZ)
+    self.assertEqual(3, unittest_pb2.TestAllTypes.BAZ)
+
+  def testContainingTypeBehaviorForExtensions(self):
+    self.assertEqual(unittest_pb2.optional_int32_extension.containing_type,
+                     unittest_pb2.TestAllExtensions.DESCRIPTOR)
+    self.assertEqual(unittest_pb2.TestRequired.single.containing_type,
+                     unittest_pb2.TestAllExtensions.DESCRIPTOR)
+
+  def testExtensionScope(self):
+    self.assertEqual(unittest_pb2.optional_int32_extension.extension_scope,
+                     None)
+    self.assertEqual(unittest_pb2.TestRequired.single.extension_scope,
+                     unittest_pb2.TestRequired.DESCRIPTOR)
+
+  def testIsExtension(self):
+    self.assertTrue(unittest_pb2.optional_int32_extension.is_extension)
+    self.assertTrue(unittest_pb2.TestRequired.single.is_extension)
+
+    message_descriptor = unittest_pb2.TestRequired.DESCRIPTOR
+    non_extension_descriptor = message_descriptor.fields_by_name['a']
+    self.assertTrue(not non_extension_descriptor.is_extension)
+
+  def testOptions(self):
+    proto = unittest_mset_pb2.TestMessageSet()
+    self.assertTrue(proto.DESCRIPTOR.GetOptions().message_set_wire_format)
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/input_stream.py b/python/google/protobuf/internal/input_stream.py
new file mode 100755
index 0000000..7bda17e
--- /dev/null
+++ b/python/google/protobuf/internal/input_stream.py
@@ -0,0 +1,338 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""InputStream is the primitive interface for reading bits from the wire.
+
+All protocol buffer deserialization can be expressed in terms of
+the InputStream primitives provided here.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import array
+import struct
+from google.protobuf import message
+from google.protobuf.internal import wire_format
+
+
+# Note that much of this code is ported from //net/proto/ProtocolBuffer, and
+# that the interface is strongly inspired by CodedInputStream from the C++
+# proto2 implementation.
+
+
+class InputStreamBuffer(object):
+
+  """Contains all logic for reading bits, and dealing with stream position.
+
+  If an InputStream method ever raises an exception, the stream is left
+  in an indeterminate state and is not safe for further use.
+  """
+
+  def __init__(self, s):
+    # What we really want is something like array('B', s), where elements we
+    # read from the array are already given to us as one-byte integers.  BUT
+    # using array() instead of buffer() would force full string copies to result
+    # from each GetSubBuffer() call.
+    #
+    # So, if the N serialized bytes of a single protocol buffer object are
+    # split evenly between 2 child messages, and so on recursively, using
+    # array('B', s) instead of buffer() would incur an additional N*logN bytes
+    # copied during deserialization.
+    #
+    # The higher constant overhead of having to ord() for every byte we read
+    # from the buffer in _ReadVarintHelper() could definitely lead to worse
+    # performance in many real-world scenarios, even if the asymptotic
+    # complexity is better.  However, our real answer is that the mythical
+    # Python/C extension module output mode for the protocol compiler will
+    # be blazing-fast and will eliminate most use of this class anyway.
+    self._buffer = buffer(s)
+    self._pos = 0
+
+  def EndOfStream(self):
+    """Returns true iff we're at the end of the stream.
+    If this returns true, then a call to any other InputStream method
+    will raise an exception.
+    """
+    return self._pos >= len(self._buffer)
+
+  def Position(self):
+    """Returns the current position in the stream, or equivalently, the
+    number of bytes read so far.
+    """
+    return self._pos
+
+  def GetSubBuffer(self, size=None):
+    """Returns a sequence-like object that represents a portion of our
+    underlying sequence.
+
+    Position 0 in the returned object corresponds to self.Position()
+    in this stream.
+
+    If size is specified, then the returned object ends after the
+    next "size" bytes in this stream.  If size is not specified,
+    then the returned object ends at the end of this stream.
+
+    We guarantee that the returned object R supports the Python buffer
+    interface (and thus that the call buffer(R) will work).
+
+    Note that the returned buffer is read-only.
+
+    The intended use for this method is for nested-message and nested-group
+    deserialization, where we want to make a recursive MergeFromString()
+    call on the portion of the original sequence that contains the serialized
+    nested message.  (And we'd like to do so without making unnecessary string
+    copies).
+
+    REQUIRES: size is nonnegative.
+    """
+    # Note that buffer() doesn't perform any actual string copy.
+    if size is None:
+      return buffer(self._buffer, self._pos)
+    else:
+      if size < 0:
+        raise message.DecodeError('Negative size %d' % size)
+      return buffer(self._buffer, self._pos, size)
+
+  def SkipBytes(self, num_bytes):
+    """Skip num_bytes bytes ahead, or go to the end of the stream, whichever
+    comes first.
+
+    REQUIRES: num_bytes is nonnegative.
+    """
+    if num_bytes < 0:
+      raise message.DecodeError('Negative num_bytes %d' % num_bytes)
+    self._pos += num_bytes
+    self._pos = min(self._pos, len(self._buffer))
+
+  def ReadBytes(self, size):
+    """Reads up to 'size' bytes from the stream, stopping early
+    only if we reach the end of the stream.  Returns the bytes read
+    as a string.
+    """
+    if size < 0:
+      raise message.DecodeError('Negative size %d' % size)
+    s = (self._buffer[self._pos : self._pos + size])
+    self._pos += len(s)  # Only advance by the number of bytes actually read.
+    return s
+
+  def ReadLittleEndian32(self):
+    """Interprets the next 4 bytes of the stream as a little-endian
+    encoded, unsiged 32-bit integer, and returns that integer.
+    """
+    try:
+      i = struct.unpack(wire_format.FORMAT_UINT32_LITTLE_ENDIAN,
+                        self._buffer[self._pos : self._pos + 4])
+      self._pos += 4
+      return i[0]  # unpack() result is a 1-element tuple.
+    except struct.error, e:
+      raise message.DecodeError(e)
+
+  def ReadLittleEndian64(self):
+    """Interprets the next 8 bytes of the stream as a little-endian
+    encoded, unsiged 64-bit integer, and returns that integer.
+    """
+    try:
+      i = struct.unpack(wire_format.FORMAT_UINT64_LITTLE_ENDIAN,
+                        self._buffer[self._pos : self._pos + 8])
+      self._pos += 8
+      return i[0]  # unpack() result is a 1-element tuple.
+    except struct.error, e:
+      raise message.DecodeError(e)
+
+  def ReadVarint32(self):
+    """Reads a varint from the stream, interprets this varint
+    as a signed, 32-bit integer, and returns the integer.
+    """
+    i = self.ReadVarint64()
+    if not wire_format.INT32_MIN <= i <= wire_format.INT32_MAX:
+      raise message.DecodeError('Value out of range for int32: %d' % i)
+    return int(i)
+
+  def ReadVarUInt32(self):
+    """Reads a varint from the stream, interprets this varint
+    as an unsigned, 32-bit integer, and returns the integer.
+    """
+    i = self.ReadVarUInt64()
+    if i > wire_format.UINT32_MAX:
+      raise message.DecodeError('Value out of range for uint32: %d' % i)
+    return i
+
+  def ReadVarint64(self):
+    """Reads a varint from the stream, interprets this varint
+    as a signed, 64-bit integer, and returns the integer.
+    """
+    i = self.ReadVarUInt64()
+    if i > wire_format.INT64_MAX:
+      i -= (1 << 64)
+    return i
+
+  def ReadVarUInt64(self):
+    """Reads a varint from the stream, interprets this varint
+    as an unsigned, 64-bit integer, and returns the integer.
+    """
+    i = self._ReadVarintHelper()
+    if not 0 <= i <= wire_format.UINT64_MAX:
+      raise message.DecodeError('Value out of range for uint64: %d' % i)
+    return i
+
+  def _ReadVarintHelper(self):
+    """Helper for the various varint-reading methods above.
+    Reads an unsigned, varint-encoded integer from the stream and
+    returns this integer.
+
+    Does no bounds checking except to ensure that we read at most as many bytes
+    as could possibly be present in a varint-encoded 64-bit number.
+    """
+    result = 0
+    shift = 0
+    while 1:
+      if shift >= 64:
+        raise message.DecodeError('Too many bytes when decoding varint.')
+      try:
+        b = ord(self._buffer[self._pos])
+      except IndexError:
+        raise message.DecodeError('Truncated varint.')
+      self._pos += 1
+      result |= ((b & 0x7f) << shift)
+      shift += 7
+      if not (b & 0x80):
+        return result
+
+
+class InputStreamArray(object):
+
+  """Contains all logic for reading bits, and dealing with stream position.
+
+  If an InputStream method ever raises an exception, the stream is left
+  in an indeterminate state and is not safe for further use.
+
+  This alternative to InputStreamBuffer is used in environments where buffer()
+  is unavailble, such as Google App Engine.
+  """
+
+  def __init__(self, s):
+    self._buffer = array.array('B', s)
+    self._pos = 0
+
+  def EndOfStream(self):
+    return self._pos >= len(self._buffer)
+
+  def Position(self):
+    return self._pos
+
+  def GetSubBuffer(self, size=None):
+    if size is None:
+      return self._buffer[self._pos : ].tostring()
+    else:
+      if size < 0:
+        raise message.DecodeError('Negative size %d' % size)
+      return self._buffer[self._pos : self._pos + size].tostring()
+
+  def SkipBytes(self, num_bytes):
+    if num_bytes < 0:
+      raise message.DecodeError('Negative num_bytes %d' % num_bytes)
+    self._pos += num_bytes
+    self._pos = min(self._pos, len(self._buffer))
+
+  def ReadBytes(self, size):
+    if size < 0:
+      raise message.DecodeError('Negative size %d' % size)
+    s = self._buffer[self._pos : self._pos + size].tostring()
+    self._pos += len(s)  # Only advance by the number of bytes actually read.
+    return s
+
+  def ReadLittleEndian32(self):
+    try:
+      i = struct.unpack(wire_format.FORMAT_UINT32_LITTLE_ENDIAN,
+                        self._buffer[self._pos : self._pos + 4])
+      self._pos += 4
+      return i[0]  # unpack() result is a 1-element tuple.
+    except struct.error, e:
+      raise message.DecodeError(e)
+
+  def ReadLittleEndian64(self):
+    try:
+      i = struct.unpack(wire_format.FORMAT_UINT64_LITTLE_ENDIAN,
+                        self._buffer[self._pos : self._pos + 8])
+      self._pos += 8
+      return i[0]  # unpack() result is a 1-element tuple.
+    except struct.error, e:
+      raise message.DecodeError(e)
+
+  def ReadVarint32(self):
+    i = self.ReadVarint64()
+    if not wire_format.INT32_MIN <= i <= wire_format.INT32_MAX:
+      raise message.DecodeError('Value out of range for int32: %d' % i)
+    return int(i)
+
+  def ReadVarUInt32(self):
+    i = self.ReadVarUInt64()
+    if i > wire_format.UINT32_MAX:
+      raise message.DecodeError('Value out of range for uint32: %d' % i)
+    return i
+
+  def ReadVarint64(self):
+    i = self.ReadVarUInt64()
+    if i > wire_format.INT64_MAX:
+      i -= (1 << 64)
+    return i
+
+  def ReadVarUInt64(self):
+    i = self._ReadVarintHelper()
+    if not 0 <= i <= wire_format.UINT64_MAX:
+      raise message.DecodeError('Value out of range for uint64: %d' % i)
+    return i
+
+  def _ReadVarintHelper(self):
+    result = 0
+    shift = 0
+    while 1:
+      if shift >= 64:
+        raise message.DecodeError('Too many bytes when decoding varint.')
+      try:
+        b = self._buffer[self._pos]
+      except IndexError:
+        raise message.DecodeError('Truncated varint.')
+      self._pos += 1
+      result |= ((b & 0x7f) << shift)
+      shift += 7
+      if not (b & 0x80):
+        return result
+
+
+try:
+  buffer('')
+  InputStream = InputStreamBuffer
+except NotImplementedError:
+  # Google App Engine: dev_appserver.py
+  InputStream = InputStreamArray
+except RuntimeError:
+  # Google App Engine: production
+  InputStream = InputStreamArray
diff --git a/python/google/protobuf/internal/input_stream_test.py b/python/google/protobuf/internal/input_stream_test.py
new file mode 100755
index 0000000..ecec7f7
--- /dev/null
+++ b/python/google/protobuf/internal/input_stream_test.py
@@ -0,0 +1,314 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.internal.input_stream."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import unittest
+from google.protobuf import message
+from google.protobuf.internal import wire_format
+from google.protobuf.internal import input_stream
+
+
+class InputStreamBufferTest(unittest.TestCase):
+
+  def setUp(self):
+    self.__original_input_stream = input_stream.InputStream
+    input_stream.InputStream = input_stream.InputStreamBuffer
+
+  def tearDown(self):
+    input_stream.InputStream = self.__original_input_stream
+
+  def testEndOfStream(self):
+    stream = input_stream.InputStream('abcd')
+    self.assertFalse(stream.EndOfStream())
+    self.assertEqual('abcd', stream.ReadBytes(10))
+    self.assertTrue(stream.EndOfStream())
+
+  def testPosition(self):
+    stream = input_stream.InputStream('abcd')
+    self.assertEqual(0, stream.Position())
+    self.assertEqual(0, stream.Position())  # No side-effects.
+    stream.ReadBytes(1)
+    self.assertEqual(1, stream.Position())
+    stream.ReadBytes(1)
+    self.assertEqual(2, stream.Position())
+    stream.ReadBytes(10)
+    self.assertEqual(4, stream.Position())  # Can't go past end of stream.
+
+  def testGetSubBuffer(self):
+    stream = input_stream.InputStream('abcd')
+    # Try leaving out the size.
+    self.assertEqual('abcd', str(stream.GetSubBuffer()))
+    stream.SkipBytes(1)
+    # GetSubBuffer() always starts at current size.
+    self.assertEqual('bcd', str(stream.GetSubBuffer()))
+    # Try 0-size.
+    self.assertEqual('', str(stream.GetSubBuffer(0)))
+    # Negative sizes should raise an error.
+    self.assertRaises(message.DecodeError, stream.GetSubBuffer, -1)
+    # Positive sizes should work as expected.
+    self.assertEqual('b', str(stream.GetSubBuffer(1)))
+    self.assertEqual('bc', str(stream.GetSubBuffer(2)))
+    # Sizes longer than remaining bytes in the buffer should
+    # return the whole remaining buffer.
+    self.assertEqual('bcd', str(stream.GetSubBuffer(1000)))
+
+  def testSkipBytes(self):
+    stream = input_stream.InputStream('')
+    # Skipping bytes when at the end of stream
+    # should have no effect.
+    stream.SkipBytes(0)
+    stream.SkipBytes(1)
+    stream.SkipBytes(2)
+    self.assertTrue(stream.EndOfStream())
+    self.assertEqual(0, stream.Position())
+
+    # Try skipping within a stream.
+    stream = input_stream.InputStream('abcd')
+    self.assertEqual(0, stream.Position())
+    stream.SkipBytes(1)
+    self.assertEqual(1, stream.Position())
+    stream.SkipBytes(10)  # Can't skip past the end.
+    self.assertEqual(4, stream.Position())
+
+    # Ensure that a negative skip raises an exception.
+    stream = input_stream.InputStream('abcd')
+    stream.SkipBytes(1)
+    self.assertRaises(message.DecodeError, stream.SkipBytes, -1)
+
+  def testReadBytes(self):
+    s = 'abcd'
+    # Also test going past the total stream length.
+    for i in range(len(s) + 10):
+      stream = input_stream.InputStream(s)
+      self.assertEqual(s[:i], stream.ReadBytes(i))
+      self.assertEqual(min(i, len(s)), stream.Position())
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadBytes, -1)
+
+  def EnsureFailureOnEmptyStream(self, input_stream_method):
+    """Helper for integer-parsing tests below.
+    Ensures that the given InputStream method raises a DecodeError
+    if called on a stream with no bytes remaining.
+    """
+    stream = input_stream.InputStream('')
+    self.assertRaises(message.DecodeError, input_stream_method, stream)
+
+  def testReadLittleEndian32(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadLittleEndian32)
+    s = ''
+    # Read 0.
+    s += '\x00\x00\x00\x00'
+    # Read 1.
+    s += '\x01\x00\x00\x00'
+    # Read a bunch of different bytes.
+    s += '\x01\x02\x03\x04'
+    # Read max unsigned 32-bit int.
+    s += '\xff\xff\xff\xff'
+    # Try a read with fewer than 4 bytes left in the stream.
+    s += '\x00\x00\x00'
+    stream = input_stream.InputStream(s)
+    self.assertEqual(0, stream.ReadLittleEndian32())
+    self.assertEqual(4, stream.Position())
+    self.assertEqual(1, stream.ReadLittleEndian32())
+    self.assertEqual(8, stream.Position())
+    self.assertEqual(0x04030201, stream.ReadLittleEndian32())
+    self.assertEqual(12, stream.Position())
+    self.assertEqual(wire_format.UINT32_MAX, stream.ReadLittleEndian32())
+    self.assertEqual(16, stream.Position())
+    self.assertRaises(message.DecodeError, stream.ReadLittleEndian32)
+
+  def testReadLittleEndian64(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadLittleEndian64)
+    s = ''
+    # Read 0.
+    s += '\x00\x00\x00\x00\x00\x00\x00\x00'
+    # Read 1.
+    s += '\x01\x00\x00\x00\x00\x00\x00\x00'
+    # Read a bunch of different bytes.
+    s += '\x01\x02\x03\x04\x05\x06\x07\x08'
+    # Read max unsigned 64-bit int.
+    s += '\xff\xff\xff\xff\xff\xff\xff\xff'
+    # Try a read with fewer than 8 bytes left in the stream.
+    s += '\x00\x00\x00'
+    stream = input_stream.InputStream(s)
+    self.assertEqual(0, stream.ReadLittleEndian64())
+    self.assertEqual(8, stream.Position())
+    self.assertEqual(1, stream.ReadLittleEndian64())
+    self.assertEqual(16, stream.Position())
+    self.assertEqual(0x0807060504030201, stream.ReadLittleEndian64())
+    self.assertEqual(24, stream.Position())
+    self.assertEqual(wire_format.UINT64_MAX, stream.ReadLittleEndian64())
+    self.assertEqual(32, stream.Position())
+    self.assertRaises(message.DecodeError, stream.ReadLittleEndian64)
+
+  def ReadVarintSuccessTestHelper(self, varints_and_ints, read_method):
+    """Helper for tests below that test successful reads of various varints.
+
+    Args:
+      varints_and_ints: Iterable of (str, integer) pairs, where the string
+        gives the wire encoding and the integer gives the value we expect
+        to be returned by the read_method upon encountering this string.
+      read_method: Unbound InputStream method that is capable of reading
+        the encoded strings provided in the first elements of varints_and_ints.
+    """
+    s = ''.join(s for s, i in varints_and_ints)
+    stream = input_stream.InputStream(s)
+    expected_pos = 0
+    self.assertEqual(expected_pos, stream.Position())
+    for s, expected_int in varints_and_ints:
+      self.assertEqual(expected_int, read_method(stream))
+      expected_pos += len(s)
+      self.assertEqual(expected_pos, stream.Position())
+
+  def testReadVarint32Success(self):
+    varints_and_ints = [
+        ('\x00', 0),
+        ('\x01', 1),
+        ('\x7f', 127),
+        ('\x80\x01', 128),
+        ('\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01', -1),
+        ('\xff\xff\xff\xff\x07', wire_format.INT32_MAX),
+        ('\x80\x80\x80\x80\xf8\xff\xff\xff\xff\x01', wire_format.INT32_MIN),
+        ]
+    self.ReadVarintSuccessTestHelper(varints_and_ints,
+                                     input_stream.InputStream.ReadVarint32)
+
+  def testReadVarint32Failure(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadVarint32)
+
+    # Try and fail to read INT32_MAX + 1.
+    s = '\x80\x80\x80\x80\x08'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarint32)
+
+    # Try and fail to read INT32_MIN - 1.
+    s = '\xfe\xff\xff\xff\xf7\xff\xff\xff\xff\x01'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarint32)
+
+    # Try and fail to read something that looks like
+    # a varint with more than 10 bytes.
+    s = '\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarint32)
+
+  def testReadVarUInt32Success(self):
+    varints_and_ints = [
+        ('\x00', 0),
+        ('\x01', 1),
+        ('\x7f', 127),
+        ('\x80\x01', 128),
+        ('\xff\xff\xff\xff\x0f', wire_format.UINT32_MAX),
+        ]
+    self.ReadVarintSuccessTestHelper(varints_and_ints,
+                                     input_stream.InputStream.ReadVarUInt32)
+
+  def testReadVarUInt32Failure(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadVarUInt32)
+    # Try and fail to read UINT32_MAX + 1
+    s = '\x80\x80\x80\x80\x10'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarUInt32)
+
+    # Try and fail to read something that looks like
+    # a varint with more than 10 bytes.
+    s = '\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarUInt32)
+
+  def testReadVarint64Success(self):
+    varints_and_ints = [
+        ('\x00', 0),
+        ('\x01', 1),
+        ('\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01', -1),
+        ('\x7f', 127),
+        ('\x80\x01', 128),
+        ('\xff\xff\xff\xff\xff\xff\xff\xff\x7f', wire_format.INT64_MAX),
+        ('\x80\x80\x80\x80\x80\x80\x80\x80\x80\x01', wire_format.INT64_MIN),
+        ]
+    self.ReadVarintSuccessTestHelper(varints_and_ints,
+                                     input_stream.InputStream.ReadVarint64)
+
+  def testReadVarint64Failure(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadVarint64)
+    # Try and fail to read something with the mythical 64th bit set.
+    s = '\x80\x80\x80\x80\x80\x80\x80\x80\x80\x02'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarint64)
+
+    # Try and fail to read something that looks like
+    # a varint with more than 10 bytes.
+    s = '\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarint64)
+
+  def testReadVarUInt64Success(self):
+    varints_and_ints = [
+        ('\x00', 0),
+        ('\x01', 1),
+        ('\x7f', 127),
+        ('\x80\x01', 128),
+        ('\x80\x80\x80\x80\x80\x80\x80\x80\x80\x01', 1 << 63),
+        ]
+    self.ReadVarintSuccessTestHelper(varints_and_ints,
+                                     input_stream.InputStream.ReadVarUInt64)
+
+  def testReadVarUInt64Failure(self):
+    self.EnsureFailureOnEmptyStream(input_stream.InputStream.ReadVarUInt64)
+    # Try and fail to read something with the mythical 64th bit set.
+    s = '\x80\x80\x80\x80\x80\x80\x80\x80\x80\x02'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarUInt64)
+
+    # Try and fail to read something that looks like
+    # a varint with more than 10 bytes.
+    s = '\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\x00'
+    stream = input_stream.InputStream(s)
+    self.assertRaises(message.DecodeError, stream.ReadVarUInt64)
+
+
+class InputStreamArrayTest(InputStreamBufferTest):
+
+  def setUp(self):
+    # Test InputStreamArray against the same tests in InputStreamBuffer
+    self.__original_input_stream = input_stream.InputStream
+    input_stream.InputStream = input_stream.InputStreamArray
+
+  def tearDown(self):
+    input_stream.InputStream = self.__original_input_stream
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/message_listener.py b/python/google/protobuf/internal/message_listener.py
new file mode 100755
index 0000000..4397895
--- /dev/null
+++ b/python/google/protobuf/internal/message_listener.py
@@ -0,0 +1,69 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Defines a listener interface for observing certain
+state transitions on Message objects.
+
+Also defines a null implementation of this interface.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+
+class MessageListener(object):
+
+  """Listens for transitions to nonempty and for invalidations of cached
+  byte sizes.  Meant to be registered via Message._SetListener().
+  """
+
+  def TransitionToNonempty(self):
+    """Called the *first* time that this message becomes nonempty.
+    Implementations are free (but not required) to call this method multiple
+    times after the message has become nonempty.
+    """
+    raise NotImplementedError
+
+  def ByteSizeDirty(self):
+    """Called *every* time the cached byte size value
+    for this object is invalidated (transitions from being
+    "clean" to "dirty").
+    """
+    raise NotImplementedError
+
+
+class NullMessageListener(object):
+
+  """No-op MessageListener implementation."""
+
+  def TransitionToNonempty(self):
+    pass
+
+  def ByteSizeDirty(self):
+    pass
diff --git a/python/google/protobuf/internal/message_test.py b/python/google/protobuf/internal/message_test.py
new file mode 100755
index 0000000..df344cf
--- /dev/null
+++ b/python/google/protobuf/internal/message_test.py
@@ -0,0 +1,53 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Tests python protocol buffers against the golden message."""
+
+__author__ = 'gps@google.com (Gregory P. Smith)'
+
+import unittest
+from google.protobuf import unittest_import_pb2
+from google.protobuf import unittest_pb2
+from google.protobuf.internal import test_util
+
+
+class MessageTest(test_util.GoldenMessageTestCase):
+
+  def testGoldenMessage(self):
+    golden_data = test_util.GoldenFile('golden_message').read()
+    golden_message = unittest_pb2.TestAllTypes()
+    golden_message.ParseFromString(golden_data)
+    self.ExpectAllFieldsSet(golden_message)
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/more_extensions.proto b/python/google/protobuf/internal/more_extensions.proto
new file mode 100644
index 0000000..e2d9701
--- /dev/null
+++ b/python/google/protobuf/internal/more_extensions.proto
@@ -0,0 +1,58 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+// http://code.google.com/p/protobuf/
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// Author: robinson@google.com (Will Robinson)
+
+
+package google.protobuf.internal;
+
+
+message TopLevelMessage {
+  optional ExtendedMessage submessage = 1;
+}
+
+
+message ExtendedMessage {
+  extensions 1 to max;
+}
+
+
+message ForeignMessage {
+  optional int32 foreign_message_int = 1;
+}
+
+
+extend ExtendedMessage {
+  optional int32 optional_int_extension = 1;
+  optional ForeignMessage optional_message_extension = 2;
+
+  repeated int32 repeated_int_extension = 3;
+  repeated ForeignMessage repeated_message_extension = 4;
+}
diff --git a/python/google/protobuf/internal/more_messages.proto b/python/google/protobuf/internal/more_messages.proto
new file mode 100644
index 0000000..c701b44
--- /dev/null
+++ b/python/google/protobuf/internal/more_messages.proto
@@ -0,0 +1,51 @@
+// Protocol Buffers - Google's data interchange format
+// Copyright 2008 Google Inc.  All rights reserved.
+// http://code.google.com/p/protobuf/
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// Author: robinson@google.com (Will Robinson)
+
+
+package google.protobuf.internal;
+
+// A message where tag numbers are listed out of order, to allow us to test our
+// canonicalization of serialized output, which should always be in tag order.
+// We also mix in some extensions for extra fun.
+message OutOfOrderFields {
+  optional   sint32 optional_sint32   =  5;
+  extensions 4 to 4;
+  optional   uint32 optional_uint32   =  3;
+  extensions 2 to 2;
+  optional    int32 optional_int32    =  1;
+};
+
+
+extend OutOfOrderFields {
+  optional   uint64 optional_uint64   =  4;
+  optional    int64 optional_int64    =  2;
+}
diff --git a/python/google/protobuf/internal/output_stream.py b/python/google/protobuf/internal/output_stream.py
new file mode 100755
index 0000000..6c2d6f6
--- /dev/null
+++ b/python/google/protobuf/internal/output_stream.py
@@ -0,0 +1,125 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""OutputStream is the primitive interface for sticking bits on the wire.
+
+All protocol buffer serialization can be expressed in terms of
+the OutputStream primitives provided here.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import array
+import struct
+from google.protobuf import message
+from google.protobuf.internal import wire_format
+
+
+
+# Note that much of this code is ported from //net/proto/ProtocolBuffer, and
+# that the interface is strongly inspired by CodedOutputStream from the C++
+# proto2 implementation.
+
+
+class OutputStream(object):
+
+  """Contains all logic for writing bits, and ToString() to get the result."""
+
+  def __init__(self):
+    self._buffer = array.array('B')
+
+  def AppendRawBytes(self, raw_bytes):
+    """Appends raw_bytes to our internal buffer."""
+    self._buffer.fromstring(raw_bytes)
+
+  def AppendLittleEndian32(self, unsigned_value):
+    """Appends an unsigned 32-bit integer to the internal buffer,
+    in little-endian byte order.
+    """
+    if not 0 <= unsigned_value <= wire_format.UINT32_MAX:
+      raise message.EncodeError(
+          'Unsigned 32-bit out of range: %d' % unsigned_value)
+    self._buffer.fromstring(struct.pack(
+        wire_format.FORMAT_UINT32_LITTLE_ENDIAN, unsigned_value))
+
+  def AppendLittleEndian64(self, unsigned_value):
+    """Appends an unsigned 64-bit integer to the internal buffer,
+    in little-endian byte order.
+    """
+    if not 0 <= unsigned_value <= wire_format.UINT64_MAX:
+      raise message.EncodeError(
+          'Unsigned 64-bit out of range: %d' % unsigned_value)
+    self._buffer.fromstring(struct.pack(
+        wire_format.FORMAT_UINT64_LITTLE_ENDIAN, unsigned_value))
+
+  def AppendVarint32(self, value):
+    """Appends a signed 32-bit integer to the internal buffer,
+    encoded as a varint.  (Note that a negative varint32 will
+    always require 10 bytes of space.)
+    """
+    if not wire_format.INT32_MIN <= value <= wire_format.INT32_MAX:
+      raise message.EncodeError('Value out of range: %d' % value)
+    self.AppendVarint64(value)
+
+  def AppendVarUInt32(self, value):
+    """Appends an unsigned 32-bit integer to the internal buffer,
+    encoded as a varint.
+    """
+    if not 0 <= value <= wire_format.UINT32_MAX:
+      raise message.EncodeError('Value out of range: %d' % value)
+    self.AppendVarUInt64(value)
+
+  def AppendVarint64(self, value):
+    """Appends a signed 64-bit integer to the internal buffer,
+    encoded as a varint.
+    """
+    if not wire_format.INT64_MIN <= value <= wire_format.INT64_MAX:
+      raise message.EncodeError('Value out of range: %d' % value)
+    if value < 0:
+      value += (1 << 64)
+    self.AppendVarUInt64(value)
+
+  def AppendVarUInt64(self, unsigned_value):
+    """Appends an unsigned 64-bit integer to the internal buffer,
+    encoded as a varint.
+    """
+    if not 0 <= unsigned_value <= wire_format.UINT64_MAX:
+      raise message.EncodeError('Value out of range: %d' % unsigned_value)
+    while True:
+      bits = unsigned_value & 0x7f
+      unsigned_value >>= 7
+      if not unsigned_value:
+        self._buffer.append(bits)
+        break
+      self._buffer.append(0x80|bits)
+
+  def ToString(self):
+    """Returns a string containing the bytes in our internal buffer."""
+    return self._buffer.tostring()
diff --git a/python/google/protobuf/internal/output_stream_test.py b/python/google/protobuf/internal/output_stream_test.py
new file mode 100755
index 0000000..df92eec
--- /dev/null
+++ b/python/google/protobuf/internal/output_stream_test.py
@@ -0,0 +1,178 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.internal.output_stream."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import unittest
+from google.protobuf import message
+from google.protobuf.internal import output_stream
+from google.protobuf.internal import wire_format
+
+
+class OutputStreamTest(unittest.TestCase):
+
+  def setUp(self):
+    self.stream = output_stream.OutputStream()
+
+  def testAppendRawBytes(self):
+    # Empty string.
+    self.stream.AppendRawBytes('')
+    self.assertEqual('', self.stream.ToString())
+
+    # Nonempty string.
+    self.stream.AppendRawBytes('abc')
+    self.assertEqual('abc', self.stream.ToString())
+
+    # Ensure that we're actually appending.
+    self.stream.AppendRawBytes('def')
+    self.assertEqual('abcdef', self.stream.ToString())
+
+  def AppendNumericTestHelper(self, append_fn, values_and_strings):
+    """For each (value, expected_string) pair in values_and_strings,
+    calls an OutputStream.Append*(value) method on an OutputStream and ensures
+    that the string written to that stream matches expected_string.
+
+    Args:
+      append_fn: Unbound OutputStream method that takes an integer or
+        long value as input.
+      values_and_strings: Iterable of (value, expected_string) pairs.
+    """
+    for conversion in (int, long):
+      for value, string in values_and_strings:
+        stream = output_stream.OutputStream()
+        expected_string = ''
+        append_fn(stream, conversion(value))
+        expected_string += string
+        self.assertEqual(expected_string, stream.ToString())
+
+  def AppendOverflowTestHelper(self, append_fn, value):
+    """Calls an OutputStream.Append*(value) method and asserts
+    that the method raises message.EncodeError.
+
+    Args:
+      append_fn: Unbound OutputStream method that takes an integer or
+        long value as input.
+      value: Value to pass to append_fn which should cause an
+        message.EncodeError.
+    """
+    stream = output_stream.OutputStream()
+    self.assertRaises(message.EncodeError, append_fn, stream, value)
+
+  def testAppendLittleEndian32(self):
+    append_fn = output_stream.OutputStream.AppendLittleEndian32
+    values_and_expected_strings = [
+        (0, '\x00\x00\x00\x00'),
+        (1, '\x01\x00\x00\x00'),
+        ((1 << 32) - 1, '\xff\xff\xff\xff'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, 1 << 32)
+    self.AppendOverflowTestHelper(append_fn, -1)
+
+  def testAppendLittleEndian64(self):
+    append_fn = output_stream.OutputStream.AppendLittleEndian64
+    values_and_expected_strings = [
+        (0, '\x00\x00\x00\x00\x00\x00\x00\x00'),
+        (1, '\x01\x00\x00\x00\x00\x00\x00\x00'),
+        ((1 << 64) - 1, '\xff\xff\xff\xff\xff\xff\xff\xff'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, 1 << 64)
+    self.AppendOverflowTestHelper(append_fn, -1)
+
+  def testAppendVarint32(self):
+    append_fn = output_stream.OutputStream.AppendVarint32
+    values_and_expected_strings = [
+        (0, '\x00'),
+        (1, '\x01'),
+        (127, '\x7f'),
+        (128, '\x80\x01'),
+        (-1, '\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01'),
+        (wire_format.INT32_MAX, '\xff\xff\xff\xff\x07'),
+        (wire_format.INT32_MIN, '\x80\x80\x80\x80\xf8\xff\xff\xff\xff\x01'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, wire_format.INT32_MAX + 1)
+    self.AppendOverflowTestHelper(append_fn, wire_format.INT32_MIN - 1)
+
+  def testAppendVarUInt32(self):
+    append_fn = output_stream.OutputStream.AppendVarUInt32
+    values_and_expected_strings = [
+        (0, '\x00'),
+        (1, '\x01'),
+        (127, '\x7f'),
+        (128, '\x80\x01'),
+        (wire_format.UINT32_MAX, '\xff\xff\xff\xff\x0f'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, -1)
+    self.AppendOverflowTestHelper(append_fn, wire_format.UINT32_MAX + 1)
+
+  def testAppendVarint64(self):
+    append_fn = output_stream.OutputStream.AppendVarint64
+    values_and_expected_strings = [
+        (0, '\x00'),
+        (1, '\x01'),
+        (127, '\x7f'),
+        (128, '\x80\x01'),
+        (-1, '\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01'),
+        (wire_format.INT64_MAX, '\xff\xff\xff\xff\xff\xff\xff\xff\x7f'),
+        (wire_format.INT64_MIN, '\x80\x80\x80\x80\x80\x80\x80\x80\x80\x01'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, wire_format.INT64_MAX + 1)
+    self.AppendOverflowTestHelper(append_fn, wire_format.INT64_MIN - 1)
+
+  def testAppendVarUInt64(self):
+    append_fn = output_stream.OutputStream.AppendVarUInt64
+    values_and_expected_strings = [
+        (0, '\x00'),
+        (1, '\x01'),
+        (127, '\x7f'),
+        (128, '\x80\x01'),
+        (wire_format.UINT64_MAX, '\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01'),
+        ]
+    self.AppendNumericTestHelper(append_fn, values_and_expected_strings)
+
+    self.AppendOverflowTestHelper(append_fn, -1)
+    self.AppendOverflowTestHelper(append_fn, wire_format.UINT64_MAX + 1)
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/reflection_test.py b/python/google/protobuf/internal/reflection_test.py
new file mode 100755
index 0000000..8610177
--- /dev/null
+++ b/python/google/protobuf/internal/reflection_test.py
@@ -0,0 +1,1976 @@
+#! /usr/bin/python
+# -*- coding: utf-8 -*-
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Unittest for reflection.py, which also indirectly tests the output of the
+pure-Python protocol compiler.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import operator
+
+import unittest
+# TODO(robinson): When we split this test in two, only some of these imports
+# will be necessary in each test.
+from google.protobuf import unittest_import_pb2
+from google.protobuf import unittest_mset_pb2
+from google.protobuf import unittest_pb2
+from google.protobuf import descriptor_pb2
+from google.protobuf import descriptor
+from google.protobuf import message
+from google.protobuf import reflection
+from google.protobuf.internal import more_extensions_pb2
+from google.protobuf.internal import more_messages_pb2
+from google.protobuf.internal import wire_format
+from google.protobuf.internal import test_util
+from google.protobuf.internal import decoder
+
+
+class ReflectionTest(unittest.TestCase):
+
+  def assertIs(self, values, others):
+    self.assertEqual(len(values), len(others))
+    for i in range(len(values)):
+      self.assertTrue(values[i] is others[i])
+
+  def testSimpleHasBits(self):
+    # Test a scalar.
+    proto = unittest_pb2.TestAllTypes()
+    self.assertTrue(not proto.HasField('optional_int32'))
+    self.assertEqual(0, proto.optional_int32)
+    # HasField() shouldn't be true if all we've done is
+    # read the default value.
+    self.assertTrue(not proto.HasField('optional_int32'))
+    proto.optional_int32 = 1
+    # Setting a value however *should* set the "has" bit.
+    self.assertTrue(proto.HasField('optional_int32'))
+    proto.ClearField('optional_int32')
+    # And clearing that value should unset the "has" bit.
+    self.assertTrue(not proto.HasField('optional_int32'))
+
+  def testHasBitsWithSinglyNestedScalar(self):
+    # Helper used to test foreign messages and groups.
+    #
+    # composite_field_name should be the name of a non-repeated
+    # composite (i.e., foreign or group) field in TestAllTypes,
+    # and scalar_field_name should be the name of an integer-valued
+    # scalar field within that composite.
+    #
+    # I never thought I'd miss C++ macros and templates so much. :(
+    # This helper is semantically just:
+    #
+    #   assert proto.composite_field.scalar_field == 0
+    #   assert not proto.composite_field.HasField('scalar_field')
+    #   assert not proto.HasField('composite_field')
+    #
+    #   proto.composite_field.scalar_field = 10
+    #   old_composite_field = proto.composite_field
+    #
+    #   assert proto.composite_field.scalar_field == 10
+    #   assert proto.composite_field.HasField('scalar_field')
+    #   assert proto.HasField('composite_field')
+    #
+    #   proto.ClearField('composite_field')
+    #
+    #   assert not proto.composite_field.HasField('scalar_field')
+    #   assert not proto.HasField('composite_field')
+    #   assert proto.composite_field.scalar_field == 0
+    #
+    #   # Now ensure that ClearField('composite_field') disconnected
+    #   # the old field object from the object tree...
+    #   assert old_composite_field is not proto.composite_field
+    #   old_composite_field.scalar_field = 20
+    #   assert not proto.composite_field.HasField('scalar_field')
+    #   assert not proto.HasField('composite_field')
+    def TestCompositeHasBits(composite_field_name, scalar_field_name):
+      proto = unittest_pb2.TestAllTypes()
+      # First, check that we can get the scalar value, and see that it's the
+      # default (0), but that proto.HasField('omposite') and
+      # proto.composite.HasField('scalar') will still return False.
+      composite_field = getattr(proto, composite_field_name)
+      original_scalar_value = getattr(composite_field, scalar_field_name)
+      self.assertEqual(0, original_scalar_value)
+      # Assert that the composite object does not "have" the scalar.
+      self.assertTrue(not composite_field.HasField(scalar_field_name))
+      # Assert that proto does not "have" the composite field.
+      self.assertTrue(not proto.HasField(composite_field_name))
+
+      # Now set the scalar within the composite field.  Ensure that the setting
+      # is reflected, and that proto.HasField('composite') and
+      # proto.composite.HasField('scalar') now both return True.
+      new_val = 20
+      setattr(composite_field, scalar_field_name, new_val)
+      self.assertEqual(new_val, getattr(composite_field, scalar_field_name))
+      # Hold on to a reference to the current composite_field object.
+      old_composite_field = composite_field
+      # Assert that the has methods now return true.
+      self.assertTrue(composite_field.HasField(scalar_field_name))
+      self.assertTrue(proto.HasField(composite_field_name))
+
+      # Now call the clear method...
+      proto.ClearField(composite_field_name)
+
+      # ...and ensure that the "has" bits are all back to False...
+      composite_field = getattr(proto, composite_field_name)
+      self.assertTrue(not composite_field.HasField(scalar_field_name))
+      self.assertTrue(not proto.HasField(composite_field_name))
+      # ...and ensure that the scalar field has returned to its default.
+      self.assertEqual(0, getattr(composite_field, scalar_field_name))
+
+      # Finally, ensure that modifications to the old composite field object
+      # don't have any effect on the parent.
+      #
+      # (NOTE that when we clear the composite field in the parent, we actually
+      # don't recursively clear down the tree.  Instead, we just disconnect the
+      # cleared composite from the tree.)
+      self.assertTrue(old_composite_field is not composite_field)
+      setattr(old_composite_field, scalar_field_name, new_val)
+      self.assertTrue(not composite_field.HasField(scalar_field_name))
+      self.assertTrue(not proto.HasField(composite_field_name))
+      self.assertEqual(0, getattr(composite_field, scalar_field_name))
+
+    # Test simple, single-level nesting when we set a scalar.
+    TestCompositeHasBits('optionalgroup', 'a')
+    TestCompositeHasBits('optional_nested_message', 'bb')
+    TestCompositeHasBits('optional_foreign_message', 'c')
+    TestCompositeHasBits('optional_import_message', 'd')
+
+  def testReferencesToNestedMessage(self):
+    proto = unittest_pb2.TestAllTypes()
+    nested = proto.optional_nested_message
+    del proto
+    # A previous version had a bug where this would raise an exception when
+    # hitting a now-dead weak reference.
+    nested.bb = 23
+
+  def testDisconnectingNestedMessageBeforeSettingField(self):
+    proto = unittest_pb2.TestAllTypes()
+    nested = proto.optional_nested_message
+    proto.ClearField('optional_nested_message')  # Should disconnect from parent
+    self.assertTrue(nested is not proto.optional_nested_message)
+    nested.bb = 23
+    self.assertTrue(not proto.HasField('optional_nested_message'))
+    self.assertEqual(0, proto.optional_nested_message.bb)
+
+  def testHasBitsWhenModifyingRepeatedFields(self):
+    # Test nesting when we add an element to a repeated field in a submessage.
+    proto = unittest_pb2.TestNestedMessageHasBits()
+    proto.optional_nested_message.nestedmessage_repeated_int32.append(5)
+    self.assertEqual(
+        [5], proto.optional_nested_message.nestedmessage_repeated_int32)
+    self.assertTrue(proto.HasField('optional_nested_message'))
+
+    # Do the same test, but with a repeated composite field within the
+    # submessage.
+    proto.ClearField('optional_nested_message')
+    self.assertTrue(not proto.HasField('optional_nested_message'))
+    proto.optional_nested_message.nestedmessage_repeated_foreignmessage.add()
+    self.assertTrue(proto.HasField('optional_nested_message'))
+
+  def testHasBitsForManyLevelsOfNesting(self):
+    # Test nesting many levels deep.
+    recursive_proto = unittest_pb2.TestMutualRecursionA()
+    self.assertTrue(not recursive_proto.HasField('bb'))
+    self.assertEqual(0, recursive_proto.bb.a.bb.a.bb.optional_int32)
+    self.assertTrue(not recursive_proto.HasField('bb'))
+    recursive_proto.bb.a.bb.a.bb.optional_int32 = 5
+    self.assertEqual(5, recursive_proto.bb.a.bb.a.bb.optional_int32)
+    self.assertTrue(recursive_proto.HasField('bb'))
+    self.assertTrue(recursive_proto.bb.HasField('a'))
+    self.assertTrue(recursive_proto.bb.a.HasField('bb'))
+    self.assertTrue(recursive_proto.bb.a.bb.HasField('a'))
+    self.assertTrue(recursive_proto.bb.a.bb.a.HasField('bb'))
+    self.assertTrue(not recursive_proto.bb.a.bb.a.bb.HasField('a'))
+    self.assertTrue(recursive_proto.bb.a.bb.a.bb.HasField('optional_int32'))
+
+  def testSingularListFields(self):
+    proto = unittest_pb2.TestAllTypes()
+    proto.optional_fixed32 = 1
+    proto.optional_int32 = 5
+    proto.optional_string = 'foo'
+    self.assertEqual(
+      [ (proto.DESCRIPTOR.fields_by_name['optional_int32'  ], 5),
+        (proto.DESCRIPTOR.fields_by_name['optional_fixed32'], 1),
+        (proto.DESCRIPTOR.fields_by_name['optional_string' ], 'foo') ],
+      proto.ListFields())
+
+  def testRepeatedListFields(self):
+    proto = unittest_pb2.TestAllTypes()
+    proto.repeated_fixed32.append(1)
+    proto.repeated_int32.append(5)
+    proto.repeated_int32.append(11)
+    proto.repeated_string.extend(['foo', 'bar'])
+    proto.repeated_string.extend([])
+    proto.repeated_string.append('baz')
+    proto.repeated_string.extend(str(x) for x in xrange(2))
+    proto.optional_int32 = 21
+    self.assertEqual(
+      [ (proto.DESCRIPTOR.fields_by_name['optional_int32'  ], 21),
+        (proto.DESCRIPTOR.fields_by_name['repeated_int32'  ], [5, 11]),
+        (proto.DESCRIPTOR.fields_by_name['repeated_fixed32'], [1]),
+        (proto.DESCRIPTOR.fields_by_name['repeated_string' ],
+          ['foo', 'bar', 'baz', '0', '1']) ],
+      proto.ListFields())
+
+  def testSingularListExtensions(self):
+    proto = unittest_pb2.TestAllExtensions()
+    proto.Extensions[unittest_pb2.optional_fixed32_extension] = 1
+    proto.Extensions[unittest_pb2.optional_int32_extension  ] = 5
+    proto.Extensions[unittest_pb2.optional_string_extension ] = 'foo'
+    self.assertEqual(
+      [ (unittest_pb2.optional_int32_extension  , 5),
+        (unittest_pb2.optional_fixed32_extension, 1),
+        (unittest_pb2.optional_string_extension , 'foo') ],
+      proto.ListFields())
+
+  def testRepeatedListExtensions(self):
+    proto = unittest_pb2.TestAllExtensions()
+    proto.Extensions[unittest_pb2.repeated_fixed32_extension].append(1)
+    proto.Extensions[unittest_pb2.repeated_int32_extension  ].append(5)
+    proto.Extensions[unittest_pb2.repeated_int32_extension  ].append(11)
+    proto.Extensions[unittest_pb2.repeated_string_extension ].append('foo')
+    proto.Extensions[unittest_pb2.repeated_string_extension ].append('bar')
+    proto.Extensions[unittest_pb2.repeated_string_extension ].append('baz')
+    proto.Extensions[unittest_pb2.optional_int32_extension  ] = 21
+    self.assertEqual(
+      [ (unittest_pb2.optional_int32_extension  , 21),
+        (unittest_pb2.repeated_int32_extension  , [5, 11]),
+        (unittest_pb2.repeated_fixed32_extension, [1]),
+        (unittest_pb2.repeated_string_extension , ['foo', 'bar', 'baz']) ],
+      proto.ListFields())
+
+  def testListFieldsAndExtensions(self):
+    proto = unittest_pb2.TestFieldOrderings()
+    test_util.SetAllFieldsAndExtensions(proto)
+    unittest_pb2.my_extension_int
+    self.assertEqual(
+      [ (proto.DESCRIPTOR.fields_by_name['my_int'   ], 1),
+        (unittest_pb2.my_extension_int               , 23),
+        (proto.DESCRIPTOR.fields_by_name['my_string'], 'foo'),
+        (unittest_pb2.my_extension_string            , 'bar'),
+        (proto.DESCRIPTOR.fields_by_name['my_float' ], 1.0) ],
+      proto.ListFields())
+
+  def testDefaultValues(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(0, proto.optional_int32)
+    self.assertEqual(0, proto.optional_int64)
+    self.assertEqual(0, proto.optional_uint32)
+    self.assertEqual(0, proto.optional_uint64)
+    self.assertEqual(0, proto.optional_sint32)
+    self.assertEqual(0, proto.optional_sint64)
+    self.assertEqual(0, proto.optional_fixed32)
+    self.assertEqual(0, proto.optional_fixed64)
+    self.assertEqual(0, proto.optional_sfixed32)
+    self.assertEqual(0, proto.optional_sfixed64)
+    self.assertEqual(0.0, proto.optional_float)
+    self.assertEqual(0.0, proto.optional_double)
+    self.assertEqual(False, proto.optional_bool)
+    self.assertEqual('', proto.optional_string)
+    self.assertEqual('', proto.optional_bytes)
+
+    self.assertEqual(41, proto.default_int32)
+    self.assertEqual(42, proto.default_int64)
+    self.assertEqual(43, proto.default_uint32)
+    self.assertEqual(44, proto.default_uint64)
+    self.assertEqual(-45, proto.default_sint32)
+    self.assertEqual(46, proto.default_sint64)
+    self.assertEqual(47, proto.default_fixed32)
+    self.assertEqual(48, proto.default_fixed64)
+    self.assertEqual(49, proto.default_sfixed32)
+    self.assertEqual(-50, proto.default_sfixed64)
+    self.assertEqual(51.5, proto.default_float)
+    self.assertEqual(52e3, proto.default_double)
+    self.assertEqual(True, proto.default_bool)
+    self.assertEqual('hello', proto.default_string)
+    self.assertEqual('world', proto.default_bytes)
+    self.assertEqual(unittest_pb2.TestAllTypes.BAR, proto.default_nested_enum)
+    self.assertEqual(unittest_pb2.FOREIGN_BAR, proto.default_foreign_enum)
+    self.assertEqual(unittest_import_pb2.IMPORT_BAR,
+                     proto.default_import_enum)
+
+    proto = unittest_pb2.TestExtremeDefaultValues()
+    self.assertEqual(u'\u1234', proto.utf8_string)
+
+  def testHasFieldWithUnknownFieldName(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertRaises(ValueError, proto.HasField, 'nonexistent_field')
+
+  def testClearFieldWithUnknownFieldName(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertRaises(ValueError, proto.ClearField, 'nonexistent_field')
+
+  def testDisallowedAssignments(self):
+    # It's illegal to assign values directly to repeated fields
+    # or to nonrepeated composite fields.  Ensure that this fails.
+    proto = unittest_pb2.TestAllTypes()
+    # Repeated fields.
+    self.assertRaises(AttributeError, setattr, proto, 'repeated_int32', 10)
+    # Lists shouldn't work, either.
+    self.assertRaises(AttributeError, setattr, proto, 'repeated_int32', [10])
+    # Composite fields.
+    self.assertRaises(AttributeError, setattr, proto,
+                      'optional_nested_message', 23)
+    # Assignment to a repeated nested message field without specifying
+    # the index in the array of nested messages.
+    self.assertRaises(AttributeError, setattr, proto.repeated_nested_message,
+                      'bb', 34)
+    # Assignment to an attribute of a repeated field.
+    self.assertRaises(AttributeError, setattr, proto.repeated_float,
+                      'some_attribute', 34)
+    # proto.nonexistent_field = 23 should fail as well.
+    self.assertRaises(AttributeError, setattr, proto, 'nonexistent_field', 23)
+
+  # TODO(robinson): Add type-safety check for enums.
+  def testSingleScalarTypeSafety(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertRaises(TypeError, setattr, proto, 'optional_int32', 1.1)
+    self.assertRaises(TypeError, setattr, proto, 'optional_int32', 'foo')
+    self.assertRaises(TypeError, setattr, proto, 'optional_string', 10)
+    self.assertRaises(TypeError, setattr, proto, 'optional_bytes', 10)
+
+  def testSingleScalarBoundsChecking(self):
+    def TestMinAndMaxIntegers(field_name, expected_min, expected_max):
+      pb = unittest_pb2.TestAllTypes()
+      setattr(pb, field_name, expected_min)
+      setattr(pb, field_name, expected_max)
+      self.assertRaises(ValueError, setattr, pb, field_name, expected_min - 1)
+      self.assertRaises(ValueError, setattr, pb, field_name, expected_max + 1)
+
+    TestMinAndMaxIntegers('optional_int32', -(1 << 31), (1 << 31) - 1)
+    TestMinAndMaxIntegers('optional_uint32', 0, 0xffffffff)
+    TestMinAndMaxIntegers('optional_int64', -(1 << 63), (1 << 63) - 1)
+    TestMinAndMaxIntegers('optional_uint64', 0, 0xffffffffffffffff)
+    TestMinAndMaxIntegers('optional_nested_enum', -(1 << 31), (1 << 31) - 1)
+
+  def testRepeatedScalarTypeSafety(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertRaises(TypeError, proto.repeated_int32.append, 1.1)
+    self.assertRaises(TypeError, proto.repeated_int32.append, 'foo')
+    self.assertRaises(TypeError, proto.repeated_string, 10)
+    self.assertRaises(TypeError, proto.repeated_bytes, 10)
+
+    proto.repeated_int32.append(10)
+    proto.repeated_int32[0] = 23
+    self.assertRaises(IndexError, proto.repeated_int32.__setitem__, 500, 23)
+    self.assertRaises(TypeError, proto.repeated_int32.__setitem__, 0, 'abc')
+
+  def testSingleScalarGettersAndSetters(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(0, proto.optional_int32)
+    proto.optional_int32 = 1
+    self.assertEqual(1, proto.optional_int32)
+    # TODO(robinson): Test all other scalar field types.
+
+  def testSingleScalarClearField(self):
+    proto = unittest_pb2.TestAllTypes()
+    # Should be allowed to clear something that's not there (a no-op).
+    proto.ClearField('optional_int32')
+    proto.optional_int32 = 1
+    self.assertTrue(proto.HasField('optional_int32'))
+    proto.ClearField('optional_int32')
+    self.assertEqual(0, proto.optional_int32)
+    self.assertTrue(not proto.HasField('optional_int32'))
+    # TODO(robinson): Test all other scalar field types.
+
+  def testEnums(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(1, proto.FOO)
+    self.assertEqual(1, unittest_pb2.TestAllTypes.FOO)
+    self.assertEqual(2, proto.BAR)
+    self.assertEqual(2, unittest_pb2.TestAllTypes.BAR)
+    self.assertEqual(3, proto.BAZ)
+    self.assertEqual(3, unittest_pb2.TestAllTypes.BAZ)
+
+  def testRepeatedScalars(self):
+    proto = unittest_pb2.TestAllTypes()
+
+    self.assertTrue(not proto.repeated_int32)
+    self.assertEqual(0, len(proto.repeated_int32))
+    proto.repeated_int32.append(5)
+    proto.repeated_int32.append(10)
+    proto.repeated_int32.append(15)
+    self.assertTrue(proto.repeated_int32)
+    self.assertEqual(3, len(proto.repeated_int32))
+
+    self.assertEqual([5, 10, 15], proto.repeated_int32)
+
+    # Test single retrieval.
+    self.assertEqual(5, proto.repeated_int32[0])
+    self.assertEqual(15, proto.repeated_int32[-1])
+    # Test out-of-bounds indices.
+    self.assertRaises(IndexError, proto.repeated_int32.__getitem__, 1234)
+    self.assertRaises(IndexError, proto.repeated_int32.__getitem__, -1234)
+    # Test incorrect types passed to __getitem__.
+    self.assertRaises(TypeError, proto.repeated_int32.__getitem__, 'foo')
+    self.assertRaises(TypeError, proto.repeated_int32.__getitem__, None)
+
+    # Test single assignment.
+    proto.repeated_int32[1] = 20
+    self.assertEqual([5, 20, 15], proto.repeated_int32)
+
+    # Test insertion.
+    proto.repeated_int32.insert(1, 25)
+    self.assertEqual([5, 25, 20, 15], proto.repeated_int32)
+
+    # Test slice retrieval.
+    proto.repeated_int32.append(30)
+    self.assertEqual([25, 20, 15], proto.repeated_int32[1:4])
+    self.assertEqual([5, 25, 20, 15, 30], proto.repeated_int32[:])
+
+    # Test slice assignment with an iterator
+    proto.repeated_int32[1:4] = (i for i in xrange(3))
+    self.assertEqual([5, 0, 1, 2, 30], proto.repeated_int32)
+
+    # Test slice assignment.
+    proto.repeated_int32[1:4] = [35, 40, 45]
+    self.assertEqual([5, 35, 40, 45, 30], proto.repeated_int32)
+
+    # Test that we can use the field as an iterator.
+    result = []
+    for i in proto.repeated_int32:
+      result.append(i)
+    self.assertEqual([5, 35, 40, 45, 30], result)
+
+    # Test single deletion.
+    del proto.repeated_int32[2]
+    self.assertEqual([5, 35, 45, 30], proto.repeated_int32)
+
+    # Test slice deletion.
+    del proto.repeated_int32[2:]
+    self.assertEqual([5, 35], proto.repeated_int32)
+
+    # Test clearing.
+    proto.ClearField('repeated_int32')
+    self.assertTrue(not proto.repeated_int32)
+    self.assertEqual(0, len(proto.repeated_int32))
+
+  def testRepeatedScalarsRemove(self):
+    proto = unittest_pb2.TestAllTypes()
+
+    self.assertTrue(not proto.repeated_int32)
+    self.assertEqual(0, len(proto.repeated_int32))
+    proto.repeated_int32.append(5)
+    proto.repeated_int32.append(10)
+    proto.repeated_int32.append(5)
+    proto.repeated_int32.append(5)
+
+    self.assertEqual(4, len(proto.repeated_int32))
+    proto.repeated_int32.remove(5)
+    self.assertEqual(3, len(proto.repeated_int32))
+    self.assertEqual(10, proto.repeated_int32[0])
+    self.assertEqual(5, proto.repeated_int32[1])
+    self.assertEqual(5, proto.repeated_int32[2])
+
+    proto.repeated_int32.remove(5)
+    self.assertEqual(2, len(proto.repeated_int32))
+    self.assertEqual(10, proto.repeated_int32[0])
+    self.assertEqual(5, proto.repeated_int32[1])
+
+    proto.repeated_int32.remove(10)
+    self.assertEqual(1, len(proto.repeated_int32))
+    self.assertEqual(5, proto.repeated_int32[0])
+
+    # Remove a non-existent element.
+    self.assertRaises(ValueError, proto.repeated_int32.remove, 123)
+
+  def testRepeatedComposites(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertTrue(not proto.repeated_nested_message)
+    self.assertEqual(0, len(proto.repeated_nested_message))
+    m0 = proto.repeated_nested_message.add()
+    m1 = proto.repeated_nested_message.add()
+    self.assertTrue(proto.repeated_nested_message)
+    self.assertEqual(2, len(proto.repeated_nested_message))
+    self.assertIs([m0, m1], proto.repeated_nested_message)
+    self.assertTrue(isinstance(m0, unittest_pb2.TestAllTypes.NestedMessage))
+
+    # Test out-of-bounds indices.
+    self.assertRaises(IndexError, proto.repeated_nested_message.__getitem__,
+                      1234)
+    self.assertRaises(IndexError, proto.repeated_nested_message.__getitem__,
+                      -1234)
+
+    # Test incorrect types passed to __getitem__.
+    self.assertRaises(TypeError, proto.repeated_nested_message.__getitem__,
+                      'foo')
+    self.assertRaises(TypeError, proto.repeated_nested_message.__getitem__,
+                      None)
+
+    # Test slice retrieval.
+    m2 = proto.repeated_nested_message.add()
+    m3 = proto.repeated_nested_message.add()
+    m4 = proto.repeated_nested_message.add()
+    self.assertIs([m1, m2, m3], proto.repeated_nested_message[1:4])
+    self.assertIs([m0, m1, m2, m3, m4], proto.repeated_nested_message[:])
+
+    # Test that we can use the field as an iterator.
+    result = []
+    for i in proto.repeated_nested_message:
+      result.append(i)
+    self.assertIs([m0, m1, m2, m3, m4], result)
+
+    # Test single deletion.
+    del proto.repeated_nested_message[2]
+    self.assertIs([m0, m1, m3, m4], proto.repeated_nested_message)
+
+    # Test slice deletion.
+    del proto.repeated_nested_message[2:]
+    self.assertIs([m0, m1], proto.repeated_nested_message)
+
+    # Test clearing.
+    proto.ClearField('repeated_nested_message')
+    self.assertTrue(not proto.repeated_nested_message)
+    self.assertEqual(0, len(proto.repeated_nested_message))
+
+  def testHandWrittenReflection(self):
+    # TODO(robinson): We probably need a better way to specify
+    # protocol types by hand.  But then again, this isn't something
+    # we expect many people to do.  Hmm.
+    FieldDescriptor = descriptor.FieldDescriptor
+    foo_field_descriptor = FieldDescriptor(
+        name='foo_field', full_name='MyProto.foo_field',
+        index=0, number=1, type=FieldDescriptor.TYPE_INT64,
+        cpp_type=FieldDescriptor.CPPTYPE_INT64,
+        label=FieldDescriptor.LABEL_OPTIONAL, default_value=0,
+        containing_type=None, message_type=None, enum_type=None,
+        is_extension=False, extension_scope=None,
+        options=descriptor_pb2.FieldOptions())
+    mydescriptor = descriptor.Descriptor(
+        name='MyProto', full_name='MyProto', filename='ignored',
+        containing_type=None, nested_types=[], enum_types=[],
+        fields=[foo_field_descriptor], extensions=[],
+        options=descriptor_pb2.MessageOptions())
+    class MyProtoClass(message.Message):
+      DESCRIPTOR = mydescriptor
+      __metaclass__ = reflection.GeneratedProtocolMessageType
+    myproto_instance = MyProtoClass()
+    self.assertEqual(0, myproto_instance.foo_field)
+    self.assertTrue(not myproto_instance.HasField('foo_field'))
+    myproto_instance.foo_field = 23
+    self.assertEqual(23, myproto_instance.foo_field)
+    self.assertTrue(myproto_instance.HasField('foo_field'))
+
+  def testTopLevelExtensionsForOptionalScalar(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.optional_int32_extension
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    self.assertEqual(0, extendee_proto.Extensions[extension])
+    # As with normal scalar fields, just doing a read doesn't actually set the
+    # "has" bit.
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    # Actually set the thing.
+    extendee_proto.Extensions[extension] = 23
+    self.assertEqual(23, extendee_proto.Extensions[extension])
+    self.assertTrue(extendee_proto.HasExtension(extension))
+    # Ensure that clearing works as well.
+    extendee_proto.ClearExtension(extension)
+    self.assertEqual(0, extendee_proto.Extensions[extension])
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+
+  def testTopLevelExtensionsForRepeatedScalar(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.repeated_string_extension
+    self.assertEqual(0, len(extendee_proto.Extensions[extension]))
+    extendee_proto.Extensions[extension].append('foo')
+    self.assertEqual(['foo'], extendee_proto.Extensions[extension])
+    string_list = extendee_proto.Extensions[extension]
+    extendee_proto.ClearExtension(extension)
+    self.assertEqual(0, len(extendee_proto.Extensions[extension]))
+    self.assertTrue(string_list is not extendee_proto.Extensions[extension])
+    # Shouldn't be allowed to do Extensions[extension] = 'a'
+    self.assertRaises(TypeError, operator.setitem, extendee_proto.Extensions,
+                      extension, 'a')
+
+  def testTopLevelExtensionsForOptionalMessage(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.optional_foreign_message_extension
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    self.assertEqual(0, extendee_proto.Extensions[extension].c)
+    # As with normal (non-extension) fields, merely reading from the
+    # thing shouldn't set the "has" bit.
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    extendee_proto.Extensions[extension].c = 23
+    self.assertEqual(23, extendee_proto.Extensions[extension].c)
+    self.assertTrue(extendee_proto.HasExtension(extension))
+    # Save a reference here.
+    foreign_message = extendee_proto.Extensions[extension]
+    extendee_proto.ClearExtension(extension)
+    self.assertTrue(foreign_message is not extendee_proto.Extensions[extension])
+    # Setting a field on foreign_message now shouldn't set
+    # any "has" bits on extendee_proto.
+    foreign_message.c = 42
+    self.assertEqual(42, foreign_message.c)
+    self.assertTrue(foreign_message.HasField('c'))
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    # Shouldn't be allowed to do Extensions[extension] = 'a'
+    self.assertRaises(TypeError, operator.setitem, extendee_proto.Extensions,
+                      extension, 'a')
+
+  def testTopLevelExtensionsForRepeatedMessage(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.repeatedgroup_extension
+    self.assertEqual(0, len(extendee_proto.Extensions[extension]))
+    group = extendee_proto.Extensions[extension].add()
+    group.a = 23
+    self.assertEqual(23, extendee_proto.Extensions[extension][0].a)
+    group.a = 42
+    self.assertEqual(42, extendee_proto.Extensions[extension][0].a)
+    group_list = extendee_proto.Extensions[extension]
+    extendee_proto.ClearExtension(extension)
+    self.assertEqual(0, len(extendee_proto.Extensions[extension]))
+    self.assertTrue(group_list is not extendee_proto.Extensions[extension])
+    # Shouldn't be allowed to do Extensions[extension] = 'a'
+    self.assertRaises(TypeError, operator.setitem, extendee_proto.Extensions,
+                      extension, 'a')
+
+  def testNestedExtensions(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.TestRequired.single
+
+    # We just test the non-repeated case.
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    required = extendee_proto.Extensions[extension]
+    self.assertEqual(0, required.a)
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+    required.a = 23
+    self.assertEqual(23, extendee_proto.Extensions[extension].a)
+    self.assertTrue(extendee_proto.HasExtension(extension))
+    extendee_proto.ClearExtension(extension)
+    self.assertTrue(required is not extendee_proto.Extensions[extension])
+    self.assertTrue(not extendee_proto.HasExtension(extension))
+
+  # If message A directly contains message B, and
+  # a.HasField('b') is currently False, then mutating any
+  # extension in B should change a.HasField('b') to True
+  # (and so on up the object tree).
+  def testHasBitsForAncestorsOfExtendedMessage(self):
+    # Optional scalar extension.
+    toplevel = more_extensions_pb2.TopLevelMessage()
+    self.assertTrue(not toplevel.HasField('submessage'))
+    self.assertEqual(0, toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_int_extension])
+    self.assertTrue(not toplevel.HasField('submessage'))
+    toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_int_extension] = 23
+    self.assertEqual(23, toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_int_extension])
+    self.assertTrue(toplevel.HasField('submessage'))
+
+    # Repeated scalar extension.
+    toplevel = more_extensions_pb2.TopLevelMessage()
+    self.assertTrue(not toplevel.HasField('submessage'))
+    self.assertEqual([], toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_int_extension])
+    self.assertTrue(not toplevel.HasField('submessage'))
+    toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_int_extension].append(23)
+    self.assertEqual([23], toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_int_extension])
+    self.assertTrue(toplevel.HasField('submessage'))
+
+    # Optional message extension.
+    toplevel = more_extensions_pb2.TopLevelMessage()
+    self.assertTrue(not toplevel.HasField('submessage'))
+    self.assertEqual(0, toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_message_extension].foreign_message_int)
+    self.assertTrue(not toplevel.HasField('submessage'))
+    toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_message_extension].foreign_message_int = 23
+    self.assertEqual(23, toplevel.submessage.Extensions[
+        more_extensions_pb2.optional_message_extension].foreign_message_int)
+    self.assertTrue(toplevel.HasField('submessage'))
+
+    # Repeated message extension.
+    toplevel = more_extensions_pb2.TopLevelMessage()
+    self.assertTrue(not toplevel.HasField('submessage'))
+    self.assertEqual(0, len(toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_message_extension]))
+    self.assertTrue(not toplevel.HasField('submessage'))
+    foreign = toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_message_extension].add()
+    self.assertTrue(foreign is toplevel.submessage.Extensions[
+        more_extensions_pb2.repeated_message_extension][0])
+    self.assertTrue(toplevel.HasField('submessage'))
+
+  def testDisconnectionAfterClearingEmptyMessage(self):
+    toplevel = more_extensions_pb2.TopLevelMessage()
+    extendee_proto = toplevel.submessage
+    extension = more_extensions_pb2.optional_message_extension
+    extension_proto = extendee_proto.Extensions[extension]
+    extendee_proto.ClearExtension(extension)
+    extension_proto.foreign_message_int = 23
+
+    self.assertTrue(not toplevel.HasField('submessage'))
+    self.assertTrue(extension_proto is not extendee_proto.Extensions[extension])
+
+  def testExtensionFailureModes(self):
+    extendee_proto = unittest_pb2.TestAllExtensions()
+
+    # Try non-extension-handle arguments to HasExtension,
+    # ClearExtension(), and Extensions[]...
+    self.assertRaises(KeyError, extendee_proto.HasExtension, 1234)
+    self.assertRaises(KeyError, extendee_proto.ClearExtension, 1234)
+    self.assertRaises(KeyError, extendee_proto.Extensions.__getitem__, 1234)
+    self.assertRaises(KeyError, extendee_proto.Extensions.__setitem__, 1234, 5)
+
+    # Try something that *is* an extension handle, just not for
+    # this message...
+    unknown_handle = more_extensions_pb2.optional_int_extension
+    self.assertRaises(KeyError, extendee_proto.HasExtension,
+                      unknown_handle)
+    self.assertRaises(KeyError, extendee_proto.ClearExtension,
+                      unknown_handle)
+    self.assertRaises(KeyError, extendee_proto.Extensions.__getitem__,
+                      unknown_handle)
+    self.assertRaises(KeyError, extendee_proto.Extensions.__setitem__,
+                      unknown_handle, 5)
+
+    # Try call HasExtension() with a valid handle, but for a
+    # *repeated* field.  (Just as with non-extension repeated
+    # fields, Has*() isn't supported for extension repeated fields).
+    self.assertRaises(KeyError, extendee_proto.HasExtension,
+                      unittest_pb2.repeated_string_extension)
+
+  def testStaticParseFrom(self):
+    proto1 = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(proto1)
+
+    string1 = proto1.SerializeToString()
+    proto2 = unittest_pb2.TestAllTypes.FromString(string1)
+
+    # Messages should be equal.
+    self.assertEqual(proto2, proto1)
+
+  def testMergeFromSingularField(self):
+    # Test merge with just a singular field.
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.optional_int32 = 1
+
+    proto2 = unittest_pb2.TestAllTypes()
+    # This shouldn't get overwritten.
+    proto2.optional_string = 'value'
+
+    proto2.MergeFrom(proto1)
+    self.assertEqual(1, proto2.optional_int32)
+    self.assertEqual('value', proto2.optional_string)
+
+  def testMergeFromRepeatedField(self):
+    # Test merge with just a repeated field.
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.repeated_int32.append(1)
+    proto1.repeated_int32.append(2)
+
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.repeated_int32.append(0)
+    proto2.MergeFrom(proto1)
+
+    self.assertEqual(0, proto2.repeated_int32[0])
+    self.assertEqual(1, proto2.repeated_int32[1])
+    self.assertEqual(2, proto2.repeated_int32[2])
+
+  def testMergeFromOptionalGroup(self):
+    # Test merge with an optional group.
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.optionalgroup.a = 12
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.MergeFrom(proto1)
+    self.assertEqual(12, proto2.optionalgroup.a)
+
+  def testMergeFromRepeatedNestedMessage(self):
+    # Test merge with a repeated nested message.
+    proto1 = unittest_pb2.TestAllTypes()
+    m = proto1.repeated_nested_message.add()
+    m.bb = 123
+    m = proto1.repeated_nested_message.add()
+    m.bb = 321
+
+    proto2 = unittest_pb2.TestAllTypes()
+    m = proto2.repeated_nested_message.add()
+    m.bb = 999
+    proto2.MergeFrom(proto1)
+    self.assertEqual(999, proto2.repeated_nested_message[0].bb)
+    self.assertEqual(123, proto2.repeated_nested_message[1].bb)
+    self.assertEqual(321, proto2.repeated_nested_message[2].bb)
+
+  def testMergeFromAllFields(self):
+    # With all fields set.
+    proto1 = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(proto1)
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.MergeFrom(proto1)
+
+    # Messages should be equal.
+    self.assertEqual(proto2, proto1)
+
+    # Serialized string should be equal too.
+    string1 = proto1.SerializeToString()
+    string2 = proto2.SerializeToString()
+    self.assertEqual(string1, string2)
+
+  def testMergeFromExtensionsSingular(self):
+    proto1 = unittest_pb2.TestAllExtensions()
+    proto1.Extensions[unittest_pb2.optional_int32_extension] = 1
+
+    proto2 = unittest_pb2.TestAllExtensions()
+    proto2.MergeFrom(proto1)
+    self.assertEqual(
+        1, proto2.Extensions[unittest_pb2.optional_int32_extension])
+
+  def testMergeFromExtensionsRepeated(self):
+    proto1 = unittest_pb2.TestAllExtensions()
+    proto1.Extensions[unittest_pb2.repeated_int32_extension].append(1)
+    proto1.Extensions[unittest_pb2.repeated_int32_extension].append(2)
+
+    proto2 = unittest_pb2.TestAllExtensions()
+    proto2.Extensions[unittest_pb2.repeated_int32_extension].append(0)
+    proto2.MergeFrom(proto1)
+    self.assertEqual(
+        3, len(proto2.Extensions[unittest_pb2.repeated_int32_extension]))
+    self.assertEqual(
+        0, proto2.Extensions[unittest_pb2.repeated_int32_extension][0])
+    self.assertEqual(
+        1, proto2.Extensions[unittest_pb2.repeated_int32_extension][1])
+    self.assertEqual(
+        2, proto2.Extensions[unittest_pb2.repeated_int32_extension][2])
+
+  def testMergeFromExtensionsNestedMessage(self):
+    proto1 = unittest_pb2.TestAllExtensions()
+    ext1 = proto1.Extensions[
+        unittest_pb2.repeated_nested_message_extension]
+    m = ext1.add()
+    m.bb = 222
+    m = ext1.add()
+    m.bb = 333
+
+    proto2 = unittest_pb2.TestAllExtensions()
+    ext2 = proto2.Extensions[
+        unittest_pb2.repeated_nested_message_extension]
+    m = ext2.add()
+    m.bb = 111
+
+    proto2.MergeFrom(proto1)
+    ext2 = proto2.Extensions[
+        unittest_pb2.repeated_nested_message_extension]
+    self.assertEqual(3, len(ext2))
+    self.assertEqual(111, ext2[0].bb)
+    self.assertEqual(222, ext2[1].bb)
+    self.assertEqual(333, ext2[2].bb)
+
+  def testCopyFromSingularField(self):
+    # Test copy with just a singular field.
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.optional_int32 = 1
+    proto1.optional_string = 'important-text'
+
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.optional_string = 'value'
+
+    proto2.CopyFrom(proto1)
+    self.assertEqual(1, proto2.optional_int32)
+    self.assertEqual('important-text', proto2.optional_string)
+
+  def testCopyFromRepeatedField(self):
+    # Test copy with a repeated field.
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.repeated_int32.append(1)
+    proto1.repeated_int32.append(2)
+
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.repeated_int32.append(0)
+    proto2.CopyFrom(proto1)
+
+    self.assertEqual(1, proto2.repeated_int32[0])
+    self.assertEqual(2, proto2.repeated_int32[1])
+
+  def testCopyFromAllFields(self):
+    # With all fields set.
+    proto1 = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(proto1)
+    proto2 = unittest_pb2.TestAllTypes()
+    proto2.CopyFrom(proto1)
+
+    # Messages should be equal.
+    self.assertEqual(proto2, proto1)
+
+    # Serialized string should be equal too.
+    string1 = proto1.SerializeToString()
+    string2 = proto2.SerializeToString()
+    self.assertEqual(string1, string2)
+
+  def testCopyFromSelf(self):
+    proto1 = unittest_pb2.TestAllTypes()
+    proto1.repeated_int32.append(1)
+    proto1.optional_int32 = 2
+    proto1.optional_string = 'important-text'
+
+    proto1.CopyFrom(proto1)
+    self.assertEqual(1, proto1.repeated_int32[0])
+    self.assertEqual(2, proto1.optional_int32)
+    self.assertEqual('important-text', proto1.optional_string)
+
+  def testClear(self):
+    proto = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(proto)
+    # Clear the message.
+    proto.Clear()
+    self.assertEquals(proto.ByteSize(), 0)
+    empty_proto = unittest_pb2.TestAllTypes()
+    self.assertEquals(proto, empty_proto)
+
+    # Test if extensions which were set are cleared.
+    proto = unittest_pb2.TestAllExtensions()
+    test_util.SetAllExtensions(proto)
+    # Clear the message.
+    proto.Clear()
+    self.assertEquals(proto.ByteSize(), 0)
+    empty_proto = unittest_pb2.TestAllExtensions()
+    self.assertEquals(proto, empty_proto)
+
+  def testIsInitialized(self):
+    # Trivial cases - all optional fields and extensions.
+    proto = unittest_pb2.TestAllTypes()
+    self.assertTrue(proto.IsInitialized())
+    proto = unittest_pb2.TestAllExtensions()
+    self.assertTrue(proto.IsInitialized())
+
+    # The case of uninitialized required fields.
+    proto = unittest_pb2.TestRequired()
+    self.assertFalse(proto.IsInitialized())
+    proto.a = proto.b = proto.c = 2
+    self.assertTrue(proto.IsInitialized())
+
+    # The case of uninitialized submessage.
+    proto = unittest_pb2.TestRequiredForeign()
+    self.assertTrue(proto.IsInitialized())
+    proto.optional_message.a = 1
+    self.assertFalse(proto.IsInitialized())
+    proto.optional_message.b = 0
+    proto.optional_message.c = 0
+    self.assertTrue(proto.IsInitialized())
+
+    # Uninitialized repeated submessage.
+    message1 = proto.repeated_message.add()
+    self.assertFalse(proto.IsInitialized())
+    message1.a = message1.b = message1.c = 0
+    self.assertTrue(proto.IsInitialized())
+
+    # Uninitialized repeated group in an extension.
+    proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.TestRequired.multi
+    message1 = proto.Extensions[extension].add()
+    message2 = proto.Extensions[extension].add()
+    self.assertFalse(proto.IsInitialized())
+    message1.a = 1
+    message1.b = 1
+    message1.c = 1
+    self.assertFalse(proto.IsInitialized())
+    message2.a = 2
+    message2.b = 2
+    message2.c = 2
+    self.assertTrue(proto.IsInitialized())
+
+    # Uninitialized nonrepeated message in an extension.
+    proto = unittest_pb2.TestAllExtensions()
+    extension = unittest_pb2.TestRequired.single
+    proto.Extensions[extension].a = 1
+    self.assertFalse(proto.IsInitialized())
+    proto.Extensions[extension].b = 2
+    proto.Extensions[extension].c = 3
+    self.assertTrue(proto.IsInitialized())
+
+  def testStringUTF8Encoding(self):
+    proto = unittest_pb2.TestAllTypes()
+
+    # Assignment of a unicode object to a field of type 'bytes' is not allowed.
+    self.assertRaises(TypeError,
+                      setattr, proto, 'optional_bytes', u'unicode object')
+
+    # Check that the default value is of python's 'unicode' type.
+    self.assertEqual(type(proto.optional_string), unicode)
+
+    proto.optional_string = unicode('Testing')
+    self.assertEqual(proto.optional_string, str('Testing'))
+
+    # Assign a value of type 'str' which can be encoded in UTF-8.
+    proto.optional_string = str('Testing')
+    self.assertEqual(proto.optional_string, unicode('Testing'))
+
+    # Values of type 'str' are also accepted as long as they can be encoded in
+    # UTF-8.
+    self.assertEqual(type(proto.optional_string), str)
+
+    # Try to assign a 'str' value which contains bytes that aren't 7-bit ASCII.
+    self.assertRaises(ValueError,
+                      setattr, proto, 'optional_string', str('a\x80a'))
+    # Assign a 'str' object which contains a UTF-8 encoded string.
+    self.assertRaises(ValueError,
+                      setattr, proto, 'optional_string', 'Тест')
+    # No exception thrown.
+    proto.optional_string = 'abc'
+
+  def testStringUTF8Serialization(self):
+    proto = unittest_mset_pb2.TestMessageSet()
+    extension_message = unittest_mset_pb2.TestMessageSetExtension2
+    extension = extension_message.message_set_extension
+
+    test_utf8 = u'Тест'
+    test_utf8_bytes = test_utf8.encode('utf-8')
+
+    # 'Test' in another language, using UTF-8 charset.
+    proto.Extensions[extension].str = test_utf8
+
+    # Serialize using the MessageSet wire format (this is specified in the
+    # .proto file).
+    serialized = proto.SerializeToString()
+
+    # Check byte size.
+    self.assertEqual(proto.ByteSize(), len(serialized))
+
+    raw = unittest_mset_pb2.RawMessageSet()
+    raw.MergeFromString(serialized)
+
+    message2 = unittest_mset_pb2.TestMessageSetExtension2()
+
+    self.assertEqual(1, len(raw.item))
+    # Check that the type_id is the same as the tag ID in the .proto file.
+    self.assertEqual(raw.item[0].type_id, 1547769)
+
+    # Check the actually bytes on the wire.
+    self.assertTrue(
+        raw.item[0].message.endswith(test_utf8_bytes))
+    message2.MergeFromString(raw.item[0].message)
+
+    self.assertEqual(type(message2.str), unicode)
+    self.assertEqual(message2.str, test_utf8)
+
+    # How about if the bytes on the wire aren't a valid UTF-8 encoded string.
+    bytes = raw.item[0].message.replace(
+        test_utf8_bytes, len(test_utf8_bytes) * '\xff')
+    self.assertRaises(UnicodeDecodeError, message2.MergeFromString, bytes)
+
+
+#  Since we had so many tests for protocol buffer equality, we broke these out
+#  into separate TestCase classes.
+
+
+class TestAllTypesEqualityTest(unittest.TestCase):
+
+  def setUp(self):
+    self.first_proto = unittest_pb2.TestAllTypes()
+    self.second_proto = unittest_pb2.TestAllTypes()
+
+  def testSelfEquality(self):
+    self.assertEqual(self.first_proto, self.first_proto)
+
+  def testEmptyProtosEqual(self):
+    self.assertEqual(self.first_proto, self.second_proto)
+
+
+class FullProtosEqualityTest(unittest.TestCase):
+
+  """Equality tests using completely-full protos as a starting point."""
+
+  def setUp(self):
+    self.first_proto = unittest_pb2.TestAllTypes()
+    self.second_proto = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(self.first_proto)
+    test_util.SetAllFields(self.second_proto)
+
+  def testNoneNotEqual(self):
+    self.assertNotEqual(self.first_proto, None)
+    self.assertNotEqual(None, self.second_proto)
+
+  def testNotEqualToOtherMessage(self):
+    third_proto = unittest_pb2.TestRequired()
+    self.assertNotEqual(self.first_proto, third_proto)
+    self.assertNotEqual(third_proto, self.second_proto)
+
+  def testAllFieldsFilledEquality(self):
+    self.assertEqual(self.first_proto, self.second_proto)
+
+  def testNonRepeatedScalar(self):
+    # Nonrepeated scalar field change should cause inequality.
+    self.first_proto.optional_int32 += 1
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    # ...as should clearing a field.
+    self.first_proto.ClearField('optional_int32')
+    self.assertNotEqual(self.first_proto, self.second_proto)
+
+  def testNonRepeatedComposite(self):
+    # Change a nonrepeated composite field.
+    self.first_proto.optional_nested_message.bb += 1
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    self.first_proto.optional_nested_message.bb -= 1
+    self.assertEqual(self.first_proto, self.second_proto)
+    # Clear a field in the nested message.
+    self.first_proto.optional_nested_message.ClearField('bb')
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    self.first_proto.optional_nested_message.bb = (
+        self.second_proto.optional_nested_message.bb)
+    self.assertEqual(self.first_proto, self.second_proto)
+    # Remove the nested message entirely.
+    self.first_proto.ClearField('optional_nested_message')
+    self.assertNotEqual(self.first_proto, self.second_proto)
+
+  def testRepeatedScalar(self):
+    # Change a repeated scalar field.
+    self.first_proto.repeated_int32.append(5)
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    self.first_proto.ClearField('repeated_int32')
+    self.assertNotEqual(self.first_proto, self.second_proto)
+
+  def testRepeatedComposite(self):
+    # Change value within a repeated composite field.
+    self.first_proto.repeated_nested_message[0].bb += 1
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    self.first_proto.repeated_nested_message[0].bb -= 1
+    self.assertEqual(self.first_proto, self.second_proto)
+    # Add a value to a repeated composite field.
+    self.first_proto.repeated_nested_message.add()
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    self.second_proto.repeated_nested_message.add()
+    self.assertEqual(self.first_proto, self.second_proto)
+
+  def testNonRepeatedScalarHasBits(self):
+    # Ensure that we test "has" bits as well as value for
+    # nonrepeated scalar field.
+    self.first_proto.ClearField('optional_int32')
+    self.second_proto.optional_int32 = 0
+    self.assertNotEqual(self.first_proto, self.second_proto)
+
+  def testNonRepeatedCompositeHasBits(self):
+    # Ensure that we test "has" bits as well as value for
+    # nonrepeated composite field.
+    self.first_proto.ClearField('optional_nested_message')
+    self.second_proto.optional_nested_message.ClearField('bb')
+    self.assertNotEqual(self.first_proto, self.second_proto)
+    # TODO(robinson): Replace next two lines with method
+    # to set the "has" bit without changing the value,
+    # if/when such a method exists.
+    self.first_proto.optional_nested_message.bb = 0
+    self.first_proto.optional_nested_message.ClearField('bb')
+    self.assertEqual(self.first_proto, self.second_proto)
+
+
+class ExtensionEqualityTest(unittest.TestCase):
+
+  def testExtensionEquality(self):
+    first_proto = unittest_pb2.TestAllExtensions()
+    second_proto = unittest_pb2.TestAllExtensions()
+    self.assertEqual(first_proto, second_proto)
+    test_util.SetAllExtensions(first_proto)
+    self.assertNotEqual(first_proto, second_proto)
+    test_util.SetAllExtensions(second_proto)
+    self.assertEqual(first_proto, second_proto)
+
+    # Ensure that we check value equality.
+    first_proto.Extensions[unittest_pb2.optional_int32_extension] += 1
+    self.assertNotEqual(first_proto, second_proto)
+    first_proto.Extensions[unittest_pb2.optional_int32_extension] -= 1
+    self.assertEqual(first_proto, second_proto)
+
+    # Ensure that we also look at "has" bits.
+    first_proto.ClearExtension(unittest_pb2.optional_int32_extension)
+    second_proto.Extensions[unittest_pb2.optional_int32_extension] = 0
+    self.assertNotEqual(first_proto, second_proto)
+    first_proto.Extensions[unittest_pb2.optional_int32_extension] = 0
+    self.assertEqual(first_proto, second_proto)
+
+    # Ensure that differences in cached values
+    # don't matter if "has" bits are both false.
+    first_proto = unittest_pb2.TestAllExtensions()
+    second_proto = unittest_pb2.TestAllExtensions()
+    self.assertEqual(
+        0, first_proto.Extensions[unittest_pb2.optional_int32_extension])
+    self.assertEqual(first_proto, second_proto)
+
+
+class MutualRecursionEqualityTest(unittest.TestCase):
+
+  def testEqualityWithMutualRecursion(self):
+    first_proto = unittest_pb2.TestMutualRecursionA()
+    second_proto = unittest_pb2.TestMutualRecursionA()
+    self.assertEqual(first_proto, second_proto)
+    first_proto.bb.a.bb.optional_int32 = 23
+    self.assertNotEqual(first_proto, second_proto)
+    second_proto.bb.a.bb.optional_int32 = 23
+    self.assertEqual(first_proto, second_proto)
+
+
+class ByteSizeTest(unittest.TestCase):
+
+  def setUp(self):
+    self.proto = unittest_pb2.TestAllTypes()
+    self.extended_proto = more_extensions_pb2.ExtendedMessage()
+    self.packed_proto = unittest_pb2.TestPackedTypes()
+    self.packed_extended_proto = unittest_pb2.TestPackedExtensions()
+
+  def Size(self):
+    return self.proto.ByteSize()
+
+  def testEmptyMessage(self):
+    self.assertEqual(0, self.proto.ByteSize())
+
+  def testVarints(self):
+    def Test(i, expected_varint_size):
+      self.proto.Clear()
+      self.proto.optional_int64 = i
+      # Add one to the varint size for the tag info
+      # for tag 1.
+      self.assertEqual(expected_varint_size + 1, self.Size())
+    Test(0, 1)
+    Test(1, 1)
+    for i, num_bytes in zip(range(7, 63, 7), range(1, 10000)):
+      Test((1 << i) - 1, num_bytes)
+    Test(-1, 10)
+    Test(-2, 10)
+    Test(-(1 << 63), 10)
+
+  def testStrings(self):
+    self.proto.optional_string = ''
+    # Need one byte for tag info (tag #14), and one byte for length.
+    self.assertEqual(2, self.Size())
+
+    self.proto.optional_string = 'abc'
+    # Need one byte for tag info (tag #14), and one byte for length.
+    self.assertEqual(2 + len(self.proto.optional_string), self.Size())
+
+    self.proto.optional_string = 'x' * 128
+    # Need one byte for tag info (tag #14), and TWO bytes for length.
+    self.assertEqual(3 + len(self.proto.optional_string), self.Size())
+
+  def testOtherNumerics(self):
+    self.proto.optional_fixed32 = 1234
+    # One byte for tag and 4 bytes for fixed32.
+    self.assertEqual(5, self.Size())
+    self.proto = unittest_pb2.TestAllTypes()
+
+    self.proto.optional_fixed64 = 1234
+    # One byte for tag and 8 bytes for fixed64.
+    self.assertEqual(9, self.Size())
+    self.proto = unittest_pb2.TestAllTypes()
+
+    self.proto.optional_float = 1.234
+    # One byte for tag and 4 bytes for float.
+    self.assertEqual(5, self.Size())
+    self.proto = unittest_pb2.TestAllTypes()
+
+    self.proto.optional_double = 1.234
+    # One byte for tag and 8 bytes for float.
+    self.assertEqual(9, self.Size())
+    self.proto = unittest_pb2.TestAllTypes()
+
+    self.proto.optional_sint32 = 64
+    # One byte for tag and 2 bytes for zig-zag-encoded 64.
+    self.assertEqual(3, self.Size())
+    self.proto = unittest_pb2.TestAllTypes()
+
+  def testComposites(self):
+    # 3 bytes.
+    self.proto.optional_nested_message.bb = (1 << 14)
+    # Plus one byte for bb tag.
+    # Plus 1 byte for optional_nested_message serialized size.
+    # Plus two bytes for optional_nested_message tag.
+    self.assertEqual(3 + 1 + 1 + 2, self.Size())
+
+  def testGroups(self):
+    # 4 bytes.
+    self.proto.optionalgroup.a = (1 << 21)
+    # Plus two bytes for |a| tag.
+    # Plus 2 * two bytes for START_GROUP and END_GROUP tags.
+    self.assertEqual(4 + 2 + 2*2, self.Size())
+
+  def testRepeatedScalars(self):
+    self.proto.repeated_int32.append(10)  # 1 byte.
+    self.proto.repeated_int32.append(128)  # 2 bytes.
+    # Also need 2 bytes for each entry for tag.
+    self.assertEqual(1 + 2 + 2*2, self.Size())
+
+  def testRepeatedScalarsExtend(self):
+    self.proto.repeated_int32.extend([10, 128])  # 3 bytes.
+    # Also need 2 bytes for each entry for tag.
+    self.assertEqual(1 + 2 + 2*2, self.Size())
+
+  def testRepeatedScalarsRemove(self):
+    self.proto.repeated_int32.append(10)  # 1 byte.
+    self.proto.repeated_int32.append(128)  # 2 bytes.
+    # Also need 2 bytes for each entry for tag.
+    self.assertEqual(1 + 2 + 2*2, self.Size())
+    self.proto.repeated_int32.remove(128)
+    self.assertEqual(1 + 2, self.Size())
+
+  def testRepeatedComposites(self):
+    # Empty message.  2 bytes tag plus 1 byte length.
+    foreign_message_0 = self.proto.repeated_nested_message.add()
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    foreign_message_1 = self.proto.repeated_nested_message.add()
+    foreign_message_1.bb = 7
+    self.assertEqual(2 + 1 + 2 + 1 + 1 + 1, self.Size())
+
+  def testRepeatedCompositesDelete(self):
+    # Empty message.  2 bytes tag plus 1 byte length.
+    foreign_message_0 = self.proto.repeated_nested_message.add()
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    foreign_message_1 = self.proto.repeated_nested_message.add()
+    foreign_message_1.bb = 9
+    self.assertEqual(2 + 1 + 2 + 1 + 1 + 1, self.Size())
+
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    del self.proto.repeated_nested_message[0]
+    self.assertEqual(2 + 1 + 1 + 1, self.Size())
+
+    # Now add a new message.
+    foreign_message_2 = self.proto.repeated_nested_message.add()
+    foreign_message_2.bb = 12
+
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    self.assertEqual(2 + 1 + 1 + 1 + 2 + 1 + 1 + 1, self.Size())
+
+    # 2 bytes tag plus 1 byte length plus 1 byte bb tag 1 byte int.
+    del self.proto.repeated_nested_message[1]
+    self.assertEqual(2 + 1 + 1 + 1, self.Size())
+
+    del self.proto.repeated_nested_message[0]
+    self.assertEqual(0, self.Size())
+
+  def testRepeatedGroups(self):
+    # 2-byte START_GROUP plus 2-byte END_GROUP.
+    group_0 = self.proto.repeatedgroup.add()
+    # 2-byte START_GROUP plus 2-byte |a| tag + 1-byte |a|
+    # plus 2-byte END_GROUP.
+    group_1 = self.proto.repeatedgroup.add()
+    group_1.a =  7
+    self.assertEqual(2 + 2 + 2 + 2 + 1 + 2, self.Size())
+
+  def testExtensions(self):
+    proto = unittest_pb2.TestAllExtensions()
+    self.assertEqual(0, proto.ByteSize())
+    extension = unittest_pb2.optional_int32_extension  # Field #1, 1 byte.
+    proto.Extensions[extension] = 23
+    # 1 byte for tag, 1 byte for value.
+    self.assertEqual(2, proto.ByteSize())
+
+  def testCacheInvalidationForNonrepeatedScalar(self):
+    # Test non-extension.
+    self.proto.optional_int32 = 1
+    self.assertEqual(2, self.proto.ByteSize())
+    self.proto.optional_int32 = 128
+    self.assertEqual(3, self.proto.ByteSize())
+    self.proto.ClearField('optional_int32')
+    self.assertEqual(0, self.proto.ByteSize())
+
+    # Test within extension.
+    extension = more_extensions_pb2.optional_int_extension
+    self.extended_proto.Extensions[extension] = 1
+    self.assertEqual(2, self.extended_proto.ByteSize())
+    self.extended_proto.Extensions[extension] = 128
+    self.assertEqual(3, self.extended_proto.ByteSize())
+    self.extended_proto.ClearExtension(extension)
+    self.assertEqual(0, self.extended_proto.ByteSize())
+
+  def testCacheInvalidationForRepeatedScalar(self):
+    # Test non-extension.
+    self.proto.repeated_int32.append(1)
+    self.assertEqual(3, self.proto.ByteSize())
+    self.proto.repeated_int32.append(1)
+    self.assertEqual(6, self.proto.ByteSize())
+    self.proto.repeated_int32[1] = 128
+    self.assertEqual(7, self.proto.ByteSize())
+    self.proto.ClearField('repeated_int32')
+    self.assertEqual(0, self.proto.ByteSize())
+
+    # Test within extension.
+    extension = more_extensions_pb2.repeated_int_extension
+    repeated = self.extended_proto.Extensions[extension]
+    repeated.append(1)
+    self.assertEqual(2, self.extended_proto.ByteSize())
+    repeated.append(1)
+    self.assertEqual(4, self.extended_proto.ByteSize())
+    repeated[1] = 128
+    self.assertEqual(5, self.extended_proto.ByteSize())
+    self.extended_proto.ClearExtension(extension)
+    self.assertEqual(0, self.extended_proto.ByteSize())
+
+  def testCacheInvalidationForNonrepeatedMessage(self):
+    # Test non-extension.
+    self.proto.optional_foreign_message.c = 1
+    self.assertEqual(5, self.proto.ByteSize())
+    self.proto.optional_foreign_message.c = 128
+    self.assertEqual(6, self.proto.ByteSize())
+    self.proto.optional_foreign_message.ClearField('c')
+    self.assertEqual(3, self.proto.ByteSize())
+    self.proto.ClearField('optional_foreign_message')
+    self.assertEqual(0, self.proto.ByteSize())
+    child = self.proto.optional_foreign_message
+    self.proto.ClearField('optional_foreign_message')
+    child.c = 128
+    self.assertEqual(0, self.proto.ByteSize())
+
+    # Test within extension.
+    extension = more_extensions_pb2.optional_message_extension
+    child = self.extended_proto.Extensions[extension]
+    self.assertEqual(0, self.extended_proto.ByteSize())
+    child.foreign_message_int = 1
+    self.assertEqual(4, self.extended_proto.ByteSize())
+    child.foreign_message_int = 128
+    self.assertEqual(5, self.extended_proto.ByteSize())
+    self.extended_proto.ClearExtension(extension)
+    self.assertEqual(0, self.extended_proto.ByteSize())
+
+  def testCacheInvalidationForRepeatedMessage(self):
+    # Test non-extension.
+    child0 = self.proto.repeated_foreign_message.add()
+    self.assertEqual(3, self.proto.ByteSize())
+    self.proto.repeated_foreign_message.add()
+    self.assertEqual(6, self.proto.ByteSize())
+    child0.c = 1
+    self.assertEqual(8, self.proto.ByteSize())
+    self.proto.ClearField('repeated_foreign_message')
+    self.assertEqual(0, self.proto.ByteSize())
+
+    # Test within extension.
+    extension = more_extensions_pb2.repeated_message_extension
+    child_list = self.extended_proto.Extensions[extension]
+    child0 = child_list.add()
+    self.assertEqual(2, self.extended_proto.ByteSize())
+    child_list.add()
+    self.assertEqual(4, self.extended_proto.ByteSize())
+    child0.foreign_message_int = 1
+    self.assertEqual(6, self.extended_proto.ByteSize())
+    child0.ClearField('foreign_message_int')
+    self.assertEqual(4, self.extended_proto.ByteSize())
+    self.extended_proto.ClearExtension(extension)
+    self.assertEqual(0, self.extended_proto.ByteSize())
+
+  def testPackedRepeatedScalars(self):
+    self.assertEqual(0, self.packed_proto.ByteSize())
+
+    self.packed_proto.packed_int32.append(10)   # 1 byte.
+    self.packed_proto.packed_int32.append(128)  # 2 bytes.
+    # The tag is 2 bytes (the field number is 90), and the varint
+    # storing the length is 1 byte.
+    int_size = 1 + 2 + 3
+    self.assertEqual(int_size, self.packed_proto.ByteSize())
+
+    self.packed_proto.packed_double.append(4.2)   # 8 bytes
+    self.packed_proto.packed_double.append(3.25)  # 8 bytes
+    # 2 more tag bytes, 1 more length byte.
+    double_size = 8 + 8 + 3
+    self.assertEqual(int_size+double_size, self.packed_proto.ByteSize())
+
+    self.packed_proto.ClearField('packed_int32')
+    self.assertEqual(double_size, self.packed_proto.ByteSize())
+
+  def testPackedExtensions(self):
+    self.assertEqual(0, self.packed_extended_proto.ByteSize())
+    extension = self.packed_extended_proto.Extensions[
+        unittest_pb2.packed_fixed32_extension]
+    extension.extend([1, 2, 3, 4])   # 16 bytes
+    # Tag is 3 bytes.
+    self.assertEqual(19, self.packed_extended_proto.ByteSize())
+
+
+# TODO(robinson): We need cross-language serialization consistency tests.
+# Issues to be sure to cover include:
+#   * Handling of unrecognized tags ("uninterpreted_bytes").
+#   * Handling of MessageSets.
+#   * Consistent ordering of tags in the wire format,
+#     including ordering between extensions and non-extension
+#     fields.
+#   * Consistent serialization of negative numbers, especially
+#     negative int32s.
+#   * Handling of empty submessages (with and without "has"
+#     bits set).
+
+class SerializationTest(unittest.TestCase):
+
+  def testSerializeEmtpyMessage(self):
+    first_proto = unittest_pb2.TestAllTypes()
+    second_proto = unittest_pb2.TestAllTypes()
+    serialized = first_proto.SerializeToString()
+    self.assertEqual(first_proto.ByteSize(), len(serialized))
+    second_proto.MergeFromString(serialized)
+    self.assertEqual(first_proto, second_proto)
+
+  def testSerializeAllFields(self):
+    first_proto = unittest_pb2.TestAllTypes()
+    second_proto = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(first_proto)
+    serialized = first_proto.SerializeToString()
+    self.assertEqual(first_proto.ByteSize(), len(serialized))
+    second_proto.MergeFromString(serialized)
+    self.assertEqual(first_proto, second_proto)
+
+  def testSerializeAllExtensions(self):
+    first_proto = unittest_pb2.TestAllExtensions()
+    second_proto = unittest_pb2.TestAllExtensions()
+    test_util.SetAllExtensions(first_proto)
+    serialized = first_proto.SerializeToString()
+    second_proto.MergeFromString(serialized)
+    self.assertEqual(first_proto, second_proto)
+
+  def testCanonicalSerializationOrder(self):
+    proto = more_messages_pb2.OutOfOrderFields()
+    # These are also their tag numbers.  Even though we're setting these in
+    # reverse-tag order AND they're listed in reverse tag-order in the .proto
+    # file, they should nonetheless be serialized in tag order.
+    proto.optional_sint32 = 5
+    proto.Extensions[more_messages_pb2.optional_uint64] = 4
+    proto.optional_uint32 = 3
+    proto.Extensions[more_messages_pb2.optional_int64] = 2
+    proto.optional_int32 = 1
+    serialized = proto.SerializeToString()
+    self.assertEqual(proto.ByteSize(), len(serialized))
+    d = decoder.Decoder(serialized)
+    ReadTag = d.ReadFieldNumberAndWireType
+    self.assertEqual((1, wire_format.WIRETYPE_VARINT), ReadTag())
+    self.assertEqual(1, d.ReadInt32())
+    self.assertEqual((2, wire_format.WIRETYPE_VARINT), ReadTag())
+    self.assertEqual(2, d.ReadInt64())
+    self.assertEqual((3, wire_format.WIRETYPE_VARINT), ReadTag())
+    self.assertEqual(3, d.ReadUInt32())
+    self.assertEqual((4, wire_format.WIRETYPE_VARINT), ReadTag())
+    self.assertEqual(4, d.ReadUInt64())
+    self.assertEqual((5, wire_format.WIRETYPE_VARINT), ReadTag())
+    self.assertEqual(5, d.ReadSInt32())
+
+  def testCanonicalSerializationOrderSameAsCpp(self):
+    # Copy of the same test we use for C++.
+    proto = unittest_pb2.TestFieldOrderings()
+    test_util.SetAllFieldsAndExtensions(proto)
+    serialized = proto.SerializeToString()
+    test_util.ExpectAllFieldsAndExtensionsInOrder(serialized)
+
+  def testMergeFromStringWhenFieldsAlreadySet(self):
+    first_proto = unittest_pb2.TestAllTypes()
+    first_proto.repeated_string.append('foobar')
+    first_proto.optional_int32 = 23
+    first_proto.optional_nested_message.bb = 42
+    serialized = first_proto.SerializeToString()
+
+    second_proto = unittest_pb2.TestAllTypes()
+    second_proto.repeated_string.append('baz')
+    second_proto.optional_int32 = 100
+    second_proto.optional_nested_message.bb = 999
+
+    second_proto.MergeFromString(serialized)
+    # Ensure that we append to repeated fields.
+    self.assertEqual(['baz', 'foobar'], list(second_proto.repeated_string))
+    # Ensure that we overwrite nonrepeatd scalars.
+    self.assertEqual(23, second_proto.optional_int32)
+    # Ensure that we recursively call MergeFromString() on
+    # submessages.
+    self.assertEqual(42, second_proto.optional_nested_message.bb)
+
+  def testMessageSetWireFormat(self):
+    proto = unittest_mset_pb2.TestMessageSet()
+    extension_message1 = unittest_mset_pb2.TestMessageSetExtension1
+    extension_message2 = unittest_mset_pb2.TestMessageSetExtension2
+    extension1 = extension_message1.message_set_extension
+    extension2 = extension_message2.message_set_extension
+    proto.Extensions[extension1].i = 123
+    proto.Extensions[extension2].str = 'foo'
+
+    # Serialize using the MessageSet wire format (this is specified in the
+    # .proto file).
+    serialized = proto.SerializeToString()
+
+    raw = unittest_mset_pb2.RawMessageSet()
+    self.assertEqual(False,
+                     raw.DESCRIPTOR.GetOptions().message_set_wire_format)
+    raw.MergeFromString(serialized)
+    self.assertEqual(2, len(raw.item))
+
+    message1 = unittest_mset_pb2.TestMessageSetExtension1()
+    message1.MergeFromString(raw.item[0].message)
+    self.assertEqual(123, message1.i)
+
+    message2 = unittest_mset_pb2.TestMessageSetExtension2()
+    message2.MergeFromString(raw.item[1].message)
+    self.assertEqual('foo', message2.str)
+
+    # Deserialize using the MessageSet wire format.
+    proto2 = unittest_mset_pb2.TestMessageSet()
+    proto2.MergeFromString(serialized)
+    self.assertEqual(123, proto2.Extensions[extension1].i)
+    self.assertEqual('foo', proto2.Extensions[extension2].str)
+
+    # Check byte size.
+    self.assertEqual(proto2.ByteSize(), len(serialized))
+    self.assertEqual(proto.ByteSize(), len(serialized))
+
+  def testMessageSetWireFormatUnknownExtension(self):
+    # Create a message using the message set wire format with an unknown
+    # message.
+    raw = unittest_mset_pb2.RawMessageSet()
+
+    # Add an item.
+    item = raw.item.add()
+    item.type_id = 1545008
+    extension_message1 = unittest_mset_pb2.TestMessageSetExtension1
+    message1 = unittest_mset_pb2.TestMessageSetExtension1()
+    message1.i = 12345
+    item.message = message1.SerializeToString()
+
+    # Add a second, unknown extension.
+    item = raw.item.add()
+    item.type_id = 1545009
+    extension_message1 = unittest_mset_pb2.TestMessageSetExtension1
+    message1 = unittest_mset_pb2.TestMessageSetExtension1()
+    message1.i = 12346
+    item.message = message1.SerializeToString()
+
+    # Add another unknown extension.
+    item = raw.item.add()
+    item.type_id = 1545010
+    message1 = unittest_mset_pb2.TestMessageSetExtension2()
+    message1.str = 'foo'
+    item.message = message1.SerializeToString()
+
+    serialized = raw.SerializeToString()
+
+    # Parse message using the message set wire format.
+    proto = unittest_mset_pb2.TestMessageSet()
+    proto.MergeFromString(serialized)
+
+    # Check that the message parsed well.
+    extension_message1 = unittest_mset_pb2.TestMessageSetExtension1
+    extension1 = extension_message1.message_set_extension
+    self.assertEquals(12345, proto.Extensions[extension1].i)
+
+  def testUnknownFields(self):
+    proto = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(proto)
+
+    serialized = proto.SerializeToString()
+
+    # The empty message should be parsable with all of the fields
+    # unknown.
+    proto2 = unittest_pb2.TestEmptyMessage()
+
+    # Parsing this message should succeed.
+    proto2.MergeFromString(serialized)
+
+    # Now test with a int64 field set.
+    proto = unittest_pb2.TestAllTypes()
+    proto.optional_int64 = 0x0fffffffffffffff
+    serialized = proto.SerializeToString()
+    # The empty message should be parsable with all of the fields
+    # unknown.
+    proto2 = unittest_pb2.TestEmptyMessage()
+    # Parsing this message should succeed.
+    proto2.MergeFromString(serialized)
+
+  def _CheckRaises(self, exc_class, callable_obj, exception):
+    """This method checks if the excpetion type and message are as expected."""
+    try:
+      callable_obj()
+    except exc_class, ex:
+      # Check if the exception message is the right one.
+      self.assertEqual(exception, str(ex))
+      return
+    else:
+      raise self.failureException('%s not raised' % str(exc_class))
+
+  def testSerializeUninitialized(self):
+    proto = unittest_pb2.TestRequired()
+    self._CheckRaises(
+        message.EncodeError,
+        proto.SerializeToString,
+        'Required field protobuf_unittest.TestRequired.a is not set.')
+    # Shouldn't raise exceptions.
+    partial = proto.SerializePartialToString()
+
+    proto.a = 1
+    self._CheckRaises(
+        message.EncodeError,
+        proto.SerializeToString,
+        'Required field protobuf_unittest.TestRequired.b is not set.')
+    # Shouldn't raise exceptions.
+    partial = proto.SerializePartialToString()
+
+    proto.b = 2
+    self._CheckRaises(
+        message.EncodeError,
+        proto.SerializeToString,
+        'Required field protobuf_unittest.TestRequired.c is not set.')
+    # Shouldn't raise exceptions.
+    partial = proto.SerializePartialToString()
+
+    proto.c = 3
+    serialized = proto.SerializeToString()
+    # Shouldn't raise exceptions.
+    partial = proto.SerializePartialToString()
+
+    proto2 = unittest_pb2.TestRequired()
+    proto2.MergeFromString(serialized)
+    self.assertEqual(1, proto2.a)
+    self.assertEqual(2, proto2.b)
+    self.assertEqual(3, proto2.c)
+    proto2.ParseFromString(partial)
+    self.assertEqual(1, proto2.a)
+    self.assertEqual(2, proto2.b)
+    self.assertEqual(3, proto2.c)
+
+  def testSerializeAllPackedFields(self):
+    first_proto = unittest_pb2.TestPackedTypes()
+    second_proto = unittest_pb2.TestPackedTypes()
+    test_util.SetAllPackedFields(first_proto)
+    serialized = first_proto.SerializeToString()
+    self.assertEqual(first_proto.ByteSize(), len(serialized))
+    bytes_read = second_proto.MergeFromString(serialized)
+    self.assertEqual(second_proto.ByteSize(), bytes_read)
+    self.assertEqual(first_proto, second_proto)
+
+  def testSerializeAllPackedExtensions(self):
+    first_proto = unittest_pb2.TestPackedExtensions()
+    second_proto = unittest_pb2.TestPackedExtensions()
+    test_util.SetAllPackedExtensions(first_proto)
+    serialized = first_proto.SerializeToString()
+    bytes_read = second_proto.MergeFromString(serialized)
+    self.assertEqual(second_proto.ByteSize(), bytes_read)
+    self.assertEqual(first_proto, second_proto)
+
+  def testMergePackedFromStringWhenSomeFieldsAlreadySet(self):
+    first_proto = unittest_pb2.TestPackedTypes()
+    first_proto.packed_int32.extend([1, 2])
+    first_proto.packed_double.append(3.0)
+    serialized = first_proto.SerializeToString()
+
+    second_proto = unittest_pb2.TestPackedTypes()
+    second_proto.packed_int32.append(3)
+    second_proto.packed_double.extend([1.0, 2.0])
+    second_proto.packed_sint32.append(4)
+
+    second_proto.MergeFromString(serialized)
+    self.assertEqual([3, 1, 2], second_proto.packed_int32)
+    self.assertEqual([1.0, 2.0, 3.0], second_proto.packed_double)
+    self.assertEqual([4], second_proto.packed_sint32)
+
+  def testPackedFieldsWireFormat(self):
+    proto = unittest_pb2.TestPackedTypes()
+    proto.packed_int32.extend([1, 2, 150, 3])  # 1 + 1 + 2 + 1 bytes
+    proto.packed_double.extend([1.0, 1000.0])  # 8 + 8 bytes
+    proto.packed_float.append(2.0)             # 4 bytes, will be before double
+    serialized = proto.SerializeToString()
+    self.assertEqual(proto.ByteSize(), len(serialized))
+    d = decoder.Decoder(serialized)
+    ReadTag = d.ReadFieldNumberAndWireType
+    self.assertEqual((90, wire_format.WIRETYPE_LENGTH_DELIMITED), ReadTag())
+    self.assertEqual(1+1+1+2, d.ReadInt32())
+    self.assertEqual(1, d.ReadInt32())
+    self.assertEqual(2, d.ReadInt32())
+    self.assertEqual(150, d.ReadInt32())
+    self.assertEqual(3, d.ReadInt32())
+    self.assertEqual((100, wire_format.WIRETYPE_LENGTH_DELIMITED), ReadTag())
+    self.assertEqual(4, d.ReadInt32())
+    self.assertEqual(2.0, d.ReadFloat())
+    self.assertEqual((101, wire_format.WIRETYPE_LENGTH_DELIMITED), ReadTag())
+    self.assertEqual(8+8, d.ReadInt32())
+    self.assertEqual(1.0, d.ReadDouble())
+    self.assertEqual(1000.0, d.ReadDouble())
+    self.assertTrue(d.EndOfStream())
+
+  def testFieldNumbers(self):
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(unittest_pb2.TestAllTypes.NestedMessage.BB_FIELD_NUMBER, 1)
+    self.assertEqual(unittest_pb2.TestAllTypes.OPTIONAL_INT32_FIELD_NUMBER, 1)
+    self.assertEqual(unittest_pb2.TestAllTypes.OPTIONALGROUP_FIELD_NUMBER, 16)
+    self.assertEqual(
+      unittest_pb2.TestAllTypes.OPTIONAL_NESTED_MESSAGE_FIELD_NUMBER, 18)
+    self.assertEqual(
+      unittest_pb2.TestAllTypes.OPTIONAL_NESTED_ENUM_FIELD_NUMBER, 21)
+    self.assertEqual(unittest_pb2.TestAllTypes.REPEATED_INT32_FIELD_NUMBER, 31)
+    self.assertEqual(unittest_pb2.TestAllTypes.REPEATEDGROUP_FIELD_NUMBER, 46)
+    self.assertEqual(
+      unittest_pb2.TestAllTypes.REPEATED_NESTED_MESSAGE_FIELD_NUMBER, 48)
+    self.assertEqual(
+      unittest_pb2.TestAllTypes.REPEATED_NESTED_ENUM_FIELD_NUMBER, 51)
+
+  def testExtensionFieldNumbers(self):
+    self.assertEqual(unittest_pb2.TestRequired.single.number, 1000)
+    self.assertEqual(unittest_pb2.TestRequired.SINGLE_FIELD_NUMBER, 1000)
+    self.assertEqual(unittest_pb2.TestRequired.multi.number, 1001)
+    self.assertEqual(unittest_pb2.TestRequired.MULTI_FIELD_NUMBER, 1001)
+    self.assertEqual(unittest_pb2.optional_int32_extension.number, 1)
+    self.assertEqual(unittest_pb2.OPTIONAL_INT32_EXTENSION_FIELD_NUMBER, 1)
+    self.assertEqual(unittest_pb2.optionalgroup_extension.number, 16)
+    self.assertEqual(unittest_pb2.OPTIONALGROUP_EXTENSION_FIELD_NUMBER, 16)
+    self.assertEqual(unittest_pb2.optional_nested_message_extension.number, 18)
+    self.assertEqual(
+      unittest_pb2.OPTIONAL_NESTED_MESSAGE_EXTENSION_FIELD_NUMBER, 18)
+    self.assertEqual(unittest_pb2.optional_nested_enum_extension.number, 21)
+    self.assertEqual(unittest_pb2.OPTIONAL_NESTED_ENUM_EXTENSION_FIELD_NUMBER,
+      21)
+    self.assertEqual(unittest_pb2.repeated_int32_extension.number, 31)
+    self.assertEqual(unittest_pb2.REPEATED_INT32_EXTENSION_FIELD_NUMBER, 31)
+    self.assertEqual(unittest_pb2.repeatedgroup_extension.number, 46)
+    self.assertEqual(unittest_pb2.REPEATEDGROUP_EXTENSION_FIELD_NUMBER, 46)
+    self.assertEqual(unittest_pb2.repeated_nested_message_extension.number, 48)
+    self.assertEqual(
+      unittest_pb2.REPEATED_NESTED_MESSAGE_EXTENSION_FIELD_NUMBER, 48)
+    self.assertEqual(unittest_pb2.repeated_nested_enum_extension.number, 51)
+    self.assertEqual(unittest_pb2.REPEATED_NESTED_ENUM_EXTENSION_FIELD_NUMBER,
+      51)
+
+  def testInitKwargs(self):
+    proto = unittest_pb2.TestAllTypes(
+        optional_int32=1,
+        optional_string='foo',
+        optional_bool=True,
+        optional_bytes='bar',
+        optional_nested_message=unittest_pb2.TestAllTypes.NestedMessage(bb=1),
+        optional_foreign_message=unittest_pb2.ForeignMessage(c=1),
+        optional_nested_enum=unittest_pb2.TestAllTypes.FOO,
+        optional_foreign_enum=unittest_pb2.FOREIGN_FOO,
+        repeated_int32=[1, 2, 3])
+    self.assertTrue(proto.IsInitialized())
+    self.assertTrue(proto.HasField('optional_int32'))
+    self.assertTrue(proto.HasField('optional_string'))
+    self.assertTrue(proto.HasField('optional_bool'))
+    self.assertTrue(proto.HasField('optional_bytes'))
+    self.assertTrue(proto.HasField('optional_nested_message'))
+    self.assertTrue(proto.HasField('optional_foreign_message'))
+    self.assertTrue(proto.HasField('optional_nested_enum'))
+    self.assertTrue(proto.HasField('optional_foreign_enum'))
+    self.assertEqual(1, proto.optional_int32)
+    self.assertEqual('foo', proto.optional_string)
+    self.assertEqual(True, proto.optional_bool)
+    self.assertEqual('bar', proto.optional_bytes)
+    self.assertEqual(1, proto.optional_nested_message.bb)
+    self.assertEqual(1, proto.optional_foreign_message.c)
+    self.assertEqual(unittest_pb2.TestAllTypes.FOO,
+                     proto.optional_nested_enum)
+    self.assertEqual(unittest_pb2.FOREIGN_FOO, proto.optional_foreign_enum)
+    self.assertEqual([1, 2, 3], proto.repeated_int32)
+
+  def testInitArgsUnknownFieldName(self):
+    def InitalizeEmptyMessageWithExtraKeywordArg():
+      unused_proto = unittest_pb2.TestEmptyMessage(unknown='unknown')
+    self._CheckRaises(ValueError,
+                      InitalizeEmptyMessageWithExtraKeywordArg,
+                      'Protocol message has no "unknown" field.')
+
+  def testInitRequiredKwargs(self):
+    proto = unittest_pb2.TestRequired(a=1, b=1, c=1)
+    self.assertTrue(proto.IsInitialized())
+    self.assertTrue(proto.HasField('a'))
+    self.assertTrue(proto.HasField('b'))
+    self.assertTrue(proto.HasField('c'))
+    self.assertTrue(not proto.HasField('dummy2'))
+    self.assertEqual(1, proto.a)
+    self.assertEqual(1, proto.b)
+    self.assertEqual(1, proto.c)
+
+  def testInitRequiredForeignKwargs(self):
+    proto = unittest_pb2.TestRequiredForeign(
+        optional_message=unittest_pb2.TestRequired(a=1, b=1, c=1))
+    self.assertTrue(proto.IsInitialized())
+    self.assertTrue(proto.HasField('optional_message'))
+    self.assertTrue(proto.optional_message.IsInitialized())
+    self.assertTrue(proto.optional_message.HasField('a'))
+    self.assertTrue(proto.optional_message.HasField('b'))
+    self.assertTrue(proto.optional_message.HasField('c'))
+    self.assertTrue(not proto.optional_message.HasField('dummy2'))
+    self.assertEqual(unittest_pb2.TestRequired(a=1, b=1, c=1),
+                     proto.optional_message)
+    self.assertEqual(1, proto.optional_message.a)
+    self.assertEqual(1, proto.optional_message.b)
+    self.assertEqual(1, proto.optional_message.c)
+
+  def testInitRepeatedKwargs(self):
+    proto = unittest_pb2.TestAllTypes(repeated_int32=[1, 2, 3])
+    self.assertTrue(proto.IsInitialized())
+    self.assertEqual(1, proto.repeated_int32[0])
+    self.assertEqual(2, proto.repeated_int32[1])
+    self.assertEqual(3, proto.repeated_int32[2])
+
+
+class OptionsTest(unittest.TestCase):
+
+  def testMessageOptions(self):
+    proto = unittest_mset_pb2.TestMessageSet()
+    self.assertEqual(True,
+                     proto.DESCRIPTOR.GetOptions().message_set_wire_format)
+    proto = unittest_pb2.TestAllTypes()
+    self.assertEqual(False,
+                     proto.DESCRIPTOR.GetOptions().message_set_wire_format)
+
+  def testPackedOptions(self):
+    proto = unittest_pb2.TestAllTypes()
+    proto.optional_int32 = 1
+    proto.optional_double = 3.0
+    for field_descriptor, _ in proto.ListFields():
+      self.assertEqual(False, field_descriptor.GetOptions().packed)
+
+    proto = unittest_pb2.TestPackedTypes()
+    proto.packed_int32.append(1)
+    proto.packed_double.append(3.0)
+    for field_descriptor, _ in proto.ListFields():
+      self.assertEqual(True, field_descriptor.GetOptions().packed)
+      self.assertEqual(reflection._FieldDescriptor.LABEL_REPEATED,
+                       field_descriptor.label)
+
+
+class UtilityTest(unittest.TestCase):
+
+  def testImergeSorted(self):
+    ImergeSorted = reflection._ImergeSorted
+    # Various types of emptiness.
+    self.assertEqual([], list(ImergeSorted()))
+    self.assertEqual([], list(ImergeSorted([])))
+    self.assertEqual([], list(ImergeSorted([], [])))
+
+    # One nonempty list.
+    self.assertEqual([1, 2, 3], list(ImergeSorted([1, 2, 3])))
+    self.assertEqual([1, 2, 3], list(ImergeSorted([1, 2, 3], [])))
+    self.assertEqual([1, 2, 3], list(ImergeSorted([], [1, 2, 3])))
+
+    # Merging some nonempty lists together.
+    self.assertEqual([1, 2, 3], list(ImergeSorted([1, 3], [2])))
+    self.assertEqual([1, 2, 3], list(ImergeSorted([1], [3], [2])))
+    self.assertEqual([1, 2, 3], list(ImergeSorted([1], [3], [2], [])))
+
+    # Elements repeated across component iterators.
+    self.assertEqual([1, 2, 2, 3, 3],
+                     list(ImergeSorted([1, 2], [3], [2, 3])))
+
+    # Elements repeated within an iterator.
+    self.assertEqual([1, 2, 2, 3, 3],
+                     list(ImergeSorted([1, 2, 2], [3], [3])))
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/service_reflection_test.py b/python/google/protobuf/internal/service_reflection_test.py
new file mode 100755
index 0000000..e04f825
--- /dev/null
+++ b/python/google/protobuf/internal/service_reflection_test.py
@@ -0,0 +1,136 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Tests for google.protobuf.internal.service_reflection."""
+
+__author__ = 'petar@google.com (Petar Petrov)'
+
+import unittest
+from google.protobuf import unittest_pb2
+from google.protobuf import service_reflection
+from google.protobuf import service
+
+
+class FooUnitTest(unittest.TestCase):
+
+  def testService(self):
+    class MockRpcChannel(service.RpcChannel):
+      def CallMethod(self, method, controller, request, response, callback):
+        self.method = method
+        self.controller = controller
+        self.request = request
+        callback(response)
+
+    class MockRpcController(service.RpcController):
+      def SetFailed(self, msg):
+        self.failure_message = msg
+
+    self.callback_response = None
+
+    class MyService(unittest_pb2.TestService):
+      pass
+
+    self.callback_response = None
+
+    def MyCallback(response):
+      self.callback_response = response
+
+    rpc_controller = MockRpcController()
+    channel = MockRpcChannel()
+    srvc = MyService()
+    srvc.Foo(rpc_controller, unittest_pb2.FooRequest(), MyCallback)
+    self.assertEqual('Method Foo not implemented.',
+                     rpc_controller.failure_message)
+    self.assertEqual(None, self.callback_response)
+
+    rpc_controller.failure_message = None
+
+    service_descriptor = unittest_pb2.TestService.GetDescriptor()
+    srvc.CallMethod(service_descriptor.methods[1], rpc_controller,
+                    unittest_pb2.BarRequest(), MyCallback)
+    self.assertEqual('Method Bar not implemented.',
+                     rpc_controller.failure_message)
+    self.assertEqual(None, self.callback_response)
+    
+    class MyServiceImpl(unittest_pb2.TestService):
+      def Foo(self, rpc_controller, request, done):
+        self.foo_called = True
+      def Bar(self, rpc_controller, request, done):
+        self.bar_called = True
+
+    srvc = MyServiceImpl()
+    rpc_controller.failure_message = None
+    srvc.Foo(rpc_controller, unittest_pb2.FooRequest(), MyCallback)
+    self.assertEqual(None, rpc_controller.failure_message)
+    self.assertEqual(True, srvc.foo_called)
+
+    rpc_controller.failure_message = None
+    srvc.CallMethod(service_descriptor.methods[1], rpc_controller,
+                    unittest_pb2.BarRequest(), MyCallback)
+    self.assertEqual(None, rpc_controller.failure_message)
+    self.assertEqual(True, srvc.bar_called)
+
+  def testServiceStub(self):
+    class MockRpcChannel(service.RpcChannel):
+      def CallMethod(self, method, controller, request,
+                     response_class, callback):
+        self.method = method
+        self.controller = controller
+        self.request = request
+        callback(response_class())
+
+    self.callback_response = None
+
+    def MyCallback(response):
+      self.callback_response = response
+
+    channel = MockRpcChannel()
+    stub = unittest_pb2.TestService_Stub(channel)
+    rpc_controller = 'controller'
+    request = 'request'
+
+    # GetDescriptor now static, still works as instance method for compatability
+    self.assertEqual(unittest_pb2.TestService_Stub.GetDescriptor(),
+                     stub.GetDescriptor())
+
+    # Invoke method.
+    stub.Foo(rpc_controller, request, MyCallback)
+
+    self.assertTrue(isinstance(self.callback_response,
+                               unittest_pb2.FooResponse))
+    self.assertEqual(request, channel.request)
+    self.assertEqual(rpc_controller, channel.controller)
+    self.assertEqual(stub.GetDescriptor().methods[0], channel.method)
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/test_util.py b/python/google/protobuf/internal/test_util.py
new file mode 100755
index 0000000..1a0da55
--- /dev/null
+++ b/python/google/protobuf/internal/test_util.py
@@ -0,0 +1,612 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Utilities for Python proto2 tests.
+
+This is intentionally modeled on C++ code in
+//net/proto2/internal/test_util.*.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import os.path
+
+import unittest
+from google.protobuf import unittest_import_pb2
+from google.protobuf import unittest_pb2
+
+
+def SetAllFields(message):
+  """Sets every field in the message to a unique value.
+
+  Args:
+    message: A unittest_pb2.TestAllTypes instance.
+  """
+
+  #
+  # Optional fields.
+  #
+
+  message.optional_int32    = 101
+  message.optional_int64    = 102
+  message.optional_uint32   = 103
+  message.optional_uint64   = 104
+  message.optional_sint32   = 105
+  message.optional_sint64   = 106
+  message.optional_fixed32  = 107
+  message.optional_fixed64  = 108
+  message.optional_sfixed32 = 109
+  message.optional_sfixed64 = 110
+  message.optional_float    = 111
+  message.optional_double   = 112
+  message.optional_bool     = True
+  # TODO(robinson): Firmly spec out and test how
+  # protos interact with unicode.  One specific example:
+  # what happens if we change the literal below to
+  # u'115'?  What *should* happen?  Still some discussion
+  # to finish with Kenton about bytes vs. strings
+  # and forcing everything to be utf8. :-/
+  message.optional_string   = '115'
+  message.optional_bytes    = '116'
+
+  message.optionalgroup.a = 117
+  message.optional_nested_message.bb = 118
+  message.optional_foreign_message.c = 119
+  message.optional_import_message.d = 120
+
+  message.optional_nested_enum = unittest_pb2.TestAllTypes.BAZ
+  message.optional_foreign_enum = unittest_pb2.FOREIGN_BAZ
+  message.optional_import_enum = unittest_import_pb2.IMPORT_BAZ
+
+  message.optional_string_piece = '124'
+  message.optional_cord = '125'
+
+  #
+  # Repeated fields.
+  #
+
+  message.repeated_int32.append(201)
+  message.repeated_int64.append(202)
+  message.repeated_uint32.append(203)
+  message.repeated_uint64.append(204)
+  message.repeated_sint32.append(205)
+  message.repeated_sint64.append(206)
+  message.repeated_fixed32.append(207)
+  message.repeated_fixed64.append(208)
+  message.repeated_sfixed32.append(209)
+  message.repeated_sfixed64.append(210)
+  message.repeated_float.append(211)
+  message.repeated_double.append(212)
+  message.repeated_bool.append(True)
+  message.repeated_string.append('215')
+  message.repeated_bytes.append('216')
+
+  message.repeatedgroup.add().a = 217
+  message.repeated_nested_message.add().bb = 218
+  message.repeated_foreign_message.add().c = 219
+  message.repeated_import_message.add().d = 220
+
+  message.repeated_nested_enum.append(unittest_pb2.TestAllTypes.BAR)
+  message.repeated_foreign_enum.append(unittest_pb2.FOREIGN_BAR)
+  message.repeated_import_enum.append(unittest_import_pb2.IMPORT_BAR)
+
+  message.repeated_string_piece.append('224')
+  message.repeated_cord.append('225')
+
+  # Add a second one of each field.
+  message.repeated_int32.append(301)
+  message.repeated_int64.append(302)
+  message.repeated_uint32.append(303)
+  message.repeated_uint64.append(304)
+  message.repeated_sint32.append(305)
+  message.repeated_sint64.append(306)
+  message.repeated_fixed32.append(307)
+  message.repeated_fixed64.append(308)
+  message.repeated_sfixed32.append(309)
+  message.repeated_sfixed64.append(310)
+  message.repeated_float.append(311)
+  message.repeated_double.append(312)
+  message.repeated_bool.append(False)
+  message.repeated_string.append('315')
+  message.repeated_bytes.append('316')
+
+  message.repeatedgroup.add().a = 317
+  message.repeated_nested_message.add().bb = 318
+  message.repeated_foreign_message.add().c = 319
+  message.repeated_import_message.add().d = 320
+
+  message.repeated_nested_enum.append(unittest_pb2.TestAllTypes.BAZ)
+  message.repeated_foreign_enum.append(unittest_pb2.FOREIGN_BAZ)
+  message.repeated_import_enum.append(unittest_import_pb2.IMPORT_BAZ)
+
+  message.repeated_string_piece.append('324')
+  message.repeated_cord.append('325')
+
+  #
+  # Fields that have defaults.
+  #
+
+  message.default_int32 = 401
+  message.default_int64 = 402
+  message.default_uint32 = 403
+  message.default_uint64 = 404
+  message.default_sint32 = 405
+  message.default_sint64 = 406
+  message.default_fixed32 = 407
+  message.default_fixed64 = 408
+  message.default_sfixed32 = 409
+  message.default_sfixed64 = 410
+  message.default_float = 411
+  message.default_double = 412
+  message.default_bool = False
+  message.default_string = '415'
+  message.default_bytes = '416'
+
+  message.default_nested_enum = unittest_pb2.TestAllTypes.FOO
+  message.default_foreign_enum = unittest_pb2.FOREIGN_FOO
+  message.default_import_enum = unittest_import_pb2.IMPORT_FOO
+
+  message.default_string_piece = '424'
+  message.default_cord = '425'
+
+
+def SetAllExtensions(message):
+  """Sets every extension in the message to a unique value.
+
+  Args:
+    message: A unittest_pb2.TestAllExtensions instance.
+  """
+
+  extensions = message.Extensions
+  pb2 = unittest_pb2
+  import_pb2 = unittest_import_pb2
+
+  #
+  # Optional fields.
+  #
+
+  extensions[pb2.optional_int32_extension] = 101
+  extensions[pb2.optional_int64_extension] = 102
+  extensions[pb2.optional_uint32_extension] = 103
+  extensions[pb2.optional_uint64_extension] = 104
+  extensions[pb2.optional_sint32_extension] = 105
+  extensions[pb2.optional_sint64_extension] = 106
+  extensions[pb2.optional_fixed32_extension] = 107
+  extensions[pb2.optional_fixed64_extension] = 108
+  extensions[pb2.optional_sfixed32_extension] = 109
+  extensions[pb2.optional_sfixed64_extension] = 110
+  extensions[pb2.optional_float_extension] = 111
+  extensions[pb2.optional_double_extension] = 112
+  extensions[pb2.optional_bool_extension] = True
+  extensions[pb2.optional_string_extension] = '115'
+  extensions[pb2.optional_bytes_extension] = '116'
+
+  extensions[pb2.optionalgroup_extension].a = 117
+  extensions[pb2.optional_nested_message_extension].bb = 118
+  extensions[pb2.optional_foreign_message_extension].c = 119
+  extensions[pb2.optional_import_message_extension].d = 120
+
+  extensions[pb2.optional_nested_enum_extension] = pb2.TestAllTypes.BAZ
+  extensions[pb2.optional_nested_enum_extension] = pb2.TestAllTypes.BAZ
+  extensions[pb2.optional_foreign_enum_extension] = pb2.FOREIGN_BAZ
+  extensions[pb2.optional_import_enum_extension] = import_pb2.IMPORT_BAZ
+
+  extensions[pb2.optional_string_piece_extension] = '124'
+  extensions[pb2.optional_cord_extension] = '125'
+
+  #
+  # Repeated fields.
+  #
+
+  extensions[pb2.repeated_int32_extension].append(201)
+  extensions[pb2.repeated_int64_extension].append(202)
+  extensions[pb2.repeated_uint32_extension].append(203)
+  extensions[pb2.repeated_uint64_extension].append(204)
+  extensions[pb2.repeated_sint32_extension].append(205)
+  extensions[pb2.repeated_sint64_extension].append(206)
+  extensions[pb2.repeated_fixed32_extension].append(207)
+  extensions[pb2.repeated_fixed64_extension].append(208)
+  extensions[pb2.repeated_sfixed32_extension].append(209)
+  extensions[pb2.repeated_sfixed64_extension].append(210)
+  extensions[pb2.repeated_float_extension].append(211)
+  extensions[pb2.repeated_double_extension].append(212)
+  extensions[pb2.repeated_bool_extension].append(True)
+  extensions[pb2.repeated_string_extension].append('215')
+  extensions[pb2.repeated_bytes_extension].append('216')
+
+  extensions[pb2.repeatedgroup_extension].add().a = 217
+  extensions[pb2.repeated_nested_message_extension].add().bb = 218
+  extensions[pb2.repeated_foreign_message_extension].add().c = 219
+  extensions[pb2.repeated_import_message_extension].add().d = 220
+
+  extensions[pb2.repeated_nested_enum_extension].append(pb2.TestAllTypes.BAR)
+  extensions[pb2.repeated_foreign_enum_extension].append(pb2.FOREIGN_BAR)
+  extensions[pb2.repeated_import_enum_extension].append(import_pb2.IMPORT_BAR)
+
+  extensions[pb2.repeated_string_piece_extension].append('224')
+  extensions[pb2.repeated_cord_extension].append('225')
+
+  # Append a second one of each field.
+  extensions[pb2.repeated_int32_extension].append(301)
+  extensions[pb2.repeated_int64_extension].append(302)
+  extensions[pb2.repeated_uint32_extension].append(303)
+  extensions[pb2.repeated_uint64_extension].append(304)
+  extensions[pb2.repeated_sint32_extension].append(305)
+  extensions[pb2.repeated_sint64_extension].append(306)
+  extensions[pb2.repeated_fixed32_extension].append(307)
+  extensions[pb2.repeated_fixed64_extension].append(308)
+  extensions[pb2.repeated_sfixed32_extension].append(309)
+  extensions[pb2.repeated_sfixed64_extension].append(310)
+  extensions[pb2.repeated_float_extension].append(311)
+  extensions[pb2.repeated_double_extension].append(312)
+  extensions[pb2.repeated_bool_extension].append(False)
+  extensions[pb2.repeated_string_extension].append('315')
+  extensions[pb2.repeated_bytes_extension].append('316')
+
+  extensions[pb2.repeatedgroup_extension].add().a = 317
+  extensions[pb2.repeated_nested_message_extension].add().bb = 318
+  extensions[pb2.repeated_foreign_message_extension].add().c = 319
+  extensions[pb2.repeated_import_message_extension].add().d = 320
+
+  extensions[pb2.repeated_nested_enum_extension].append(pb2.TestAllTypes.BAZ)
+  extensions[pb2.repeated_foreign_enum_extension].append(pb2.FOREIGN_BAZ)
+  extensions[pb2.repeated_import_enum_extension].append(import_pb2.IMPORT_BAZ)
+
+  extensions[pb2.repeated_string_piece_extension].append('324')
+  extensions[pb2.repeated_cord_extension].append('325')
+
+  #
+  # Fields with defaults.
+  #
+
+  extensions[pb2.default_int32_extension] = 401
+  extensions[pb2.default_int64_extension] = 402
+  extensions[pb2.default_uint32_extension] = 403
+  extensions[pb2.default_uint64_extension] = 404
+  extensions[pb2.default_sint32_extension] = 405
+  extensions[pb2.default_sint64_extension] = 406
+  extensions[pb2.default_fixed32_extension] = 407
+  extensions[pb2.default_fixed64_extension] = 408
+  extensions[pb2.default_sfixed32_extension] = 409
+  extensions[pb2.default_sfixed64_extension] = 410
+  extensions[pb2.default_float_extension] = 411
+  extensions[pb2.default_double_extension] = 412
+  extensions[pb2.default_bool_extension] = False
+  extensions[pb2.default_string_extension] = '415'
+  extensions[pb2.default_bytes_extension] = '416'
+
+  extensions[pb2.default_nested_enum_extension] = pb2.TestAllTypes.FOO
+  extensions[pb2.default_foreign_enum_extension] = pb2.FOREIGN_FOO
+  extensions[pb2.default_import_enum_extension] = import_pb2.IMPORT_FOO
+
+  extensions[pb2.default_string_piece_extension] = '424'
+  extensions[pb2.default_cord_extension] = '425'
+
+
+def SetAllFieldsAndExtensions(message):
+  """Sets every field and extension in the message to a unique value.
+
+  Args:
+    message: A unittest_pb2.TestAllExtensions message.
+  """
+  message.my_int = 1
+  message.my_string = 'foo'
+  message.my_float = 1.0
+  message.Extensions[unittest_pb2.my_extension_int] = 23
+  message.Extensions[unittest_pb2.my_extension_string] = 'bar'
+
+
+def ExpectAllFieldsAndExtensionsInOrder(serialized):
+  """Ensures that serialized is the serialization we expect for a message
+  filled with SetAllFieldsAndExtensions().  (Specifically, ensures that the
+  serialization is in canonical, tag-number order).
+  """
+  my_extension_int = unittest_pb2.my_extension_int
+  my_extension_string = unittest_pb2.my_extension_string
+  expected_strings = []
+  message = unittest_pb2.TestFieldOrderings()
+  message.my_int = 1  # Field 1.
+  expected_strings.append(message.SerializeToString())
+  message.Clear()
+  message.Extensions[my_extension_int] = 23  # Field 5.
+  expected_strings.append(message.SerializeToString())
+  message.Clear()
+  message.my_string = 'foo'  # Field 11.
+  expected_strings.append(message.SerializeToString())
+  message.Clear()
+  message.Extensions[my_extension_string] = 'bar'  # Field 50.
+  expected_strings.append(message.SerializeToString())
+  message.Clear()
+  message.my_float = 1.0
+  expected_strings.append(message.SerializeToString())
+  message.Clear()
+  expected = ''.join(expected_strings)
+
+  if expected != serialized:
+    raise ValueError('Expected %r, found %r' % (expected, serialized))
+
+
+class GoldenMessageTestCase(unittest.TestCase):
+  """This adds methods to TestCase useful for verifying our Golden Message."""
+
+  def ExpectAllFieldsSet(self, message):
+    """Check all fields for correct values have after Set*Fields() is called."""
+    self.assertTrue(message.HasField('optional_int32'))
+    self.assertTrue(message.HasField('optional_int64'))
+    self.assertTrue(message.HasField('optional_uint32'))
+    self.assertTrue(message.HasField('optional_uint64'))
+    self.assertTrue(message.HasField('optional_sint32'))
+    self.assertTrue(message.HasField('optional_sint64'))
+    self.assertTrue(message.HasField('optional_fixed32'))
+    self.assertTrue(message.HasField('optional_fixed64'))
+    self.assertTrue(message.HasField('optional_sfixed32'))
+    self.assertTrue(message.HasField('optional_sfixed64'))
+    self.assertTrue(message.HasField('optional_float'))
+    self.assertTrue(message.HasField('optional_double'))
+    self.assertTrue(message.HasField('optional_bool'))
+    self.assertTrue(message.HasField('optional_string'))
+    self.assertTrue(message.HasField('optional_bytes'))
+
+    self.assertTrue(message.HasField('optionalgroup'))
+    self.assertTrue(message.HasField('optional_nested_message'))
+    self.assertTrue(message.HasField('optional_foreign_message'))
+    self.assertTrue(message.HasField('optional_import_message'))
+
+    self.assertTrue(message.optionalgroup.HasField('a'))
+    self.assertTrue(message.optional_nested_message.HasField('bb'))
+    self.assertTrue(message.optional_foreign_message.HasField('c'))
+    self.assertTrue(message.optional_import_message.HasField('d'))
+
+    self.assertTrue(message.HasField('optional_nested_enum'))
+    self.assertTrue(message.HasField('optional_foreign_enum'))
+    self.assertTrue(message.HasField('optional_import_enum'))
+
+    self.assertTrue(message.HasField('optional_string_piece'))
+    self.assertTrue(message.HasField('optional_cord'))
+
+    self.assertEqual(101, message.optional_int32)
+    self.assertEqual(102, message.optional_int64)
+    self.assertEqual(103, message.optional_uint32)
+    self.assertEqual(104, message.optional_uint64)
+    self.assertEqual(105, message.optional_sint32)
+    self.assertEqual(106, message.optional_sint64)
+    self.assertEqual(107, message.optional_fixed32)
+    self.assertEqual(108, message.optional_fixed64)
+    self.assertEqual(109, message.optional_sfixed32)
+    self.assertEqual(110, message.optional_sfixed64)
+    self.assertEqual(111, message.optional_float)
+    self.assertEqual(112, message.optional_double)
+    self.assertEqual(True, message.optional_bool)
+    self.assertEqual('115', message.optional_string)
+    self.assertEqual('116', message.optional_bytes)
+
+    self.assertEqual(117, message.optionalgroup.a);
+    self.assertEqual(118, message.optional_nested_message.bb)
+    self.assertEqual(119, message.optional_foreign_message.c)
+    self.assertEqual(120, message.optional_import_message.d)
+
+    self.assertEqual(unittest_pb2.TestAllTypes.BAZ,
+                     message.optional_nested_enum)
+    self.assertEqual(unittest_pb2.FOREIGN_BAZ, message.optional_foreign_enum)
+    self.assertEqual(unittest_import_pb2.IMPORT_BAZ,
+                     message.optional_import_enum)
+
+    # -----------------------------------------------------------------
+
+    self.assertEqual(2, len(message.repeated_int32))
+    self.assertEqual(2, len(message.repeated_int64))
+    self.assertEqual(2, len(message.repeated_uint32))
+    self.assertEqual(2, len(message.repeated_uint64))
+    self.assertEqual(2, len(message.repeated_sint32))
+    self.assertEqual(2, len(message.repeated_sint64))
+    self.assertEqual(2, len(message.repeated_fixed32))
+    self.assertEqual(2, len(message.repeated_fixed64))
+    self.assertEqual(2, len(message.repeated_sfixed32))
+    self.assertEqual(2, len(message.repeated_sfixed64))
+    self.assertEqual(2, len(message.repeated_float))
+    self.assertEqual(2, len(message.repeated_double))
+    self.assertEqual(2, len(message.repeated_bool))
+    self.assertEqual(2, len(message.repeated_string))
+    self.assertEqual(2, len(message.repeated_bytes))
+
+    self.assertEqual(2, len(message.repeatedgroup))
+    self.assertEqual(2, len(message.repeated_nested_message))
+    self.assertEqual(2, len(message.repeated_foreign_message))
+    self.assertEqual(2, len(message.repeated_import_message))
+    self.assertEqual(2, len(message.repeated_nested_enum))
+    self.assertEqual(2, len(message.repeated_foreign_enum))
+    self.assertEqual(2, len(message.repeated_import_enum))
+
+    self.assertEqual(2, len(message.repeated_string_piece))
+    self.assertEqual(2, len(message.repeated_cord))
+
+    self.assertEqual(201, message.repeated_int32[0])
+    self.assertEqual(202, message.repeated_int64[0])
+    self.assertEqual(203, message.repeated_uint32[0])
+    self.assertEqual(204, message.repeated_uint64[0])
+    self.assertEqual(205, message.repeated_sint32[0])
+    self.assertEqual(206, message.repeated_sint64[0])
+    self.assertEqual(207, message.repeated_fixed32[0])
+    self.assertEqual(208, message.repeated_fixed64[0])
+    self.assertEqual(209, message.repeated_sfixed32[0])
+    self.assertEqual(210, message.repeated_sfixed64[0])
+    self.assertEqual(211, message.repeated_float[0])
+    self.assertEqual(212, message.repeated_double[0])
+    self.assertEqual(True, message.repeated_bool[0])
+    self.assertEqual('215', message.repeated_string[0])
+    self.assertEqual('216', message.repeated_bytes[0])
+
+    self.assertEqual(217, message.repeatedgroup[0].a)
+    self.assertEqual(218, message.repeated_nested_message[0].bb)
+    self.assertEqual(219, message.repeated_foreign_message[0].c)
+    self.assertEqual(220, message.repeated_import_message[0].d)
+
+    self.assertEqual(unittest_pb2.TestAllTypes.BAR,
+                     message.repeated_nested_enum[0])
+    self.assertEqual(unittest_pb2.FOREIGN_BAR,
+                     message.repeated_foreign_enum[0])
+    self.assertEqual(unittest_import_pb2.IMPORT_BAR,
+                     message.repeated_import_enum[0])
+
+    self.assertEqual(301, message.repeated_int32[1])
+    self.assertEqual(302, message.repeated_int64[1])
+    self.assertEqual(303, message.repeated_uint32[1])
+    self.assertEqual(304, message.repeated_uint64[1])
+    self.assertEqual(305, message.repeated_sint32[1])
+    self.assertEqual(306, message.repeated_sint64[1])
+    self.assertEqual(307, message.repeated_fixed32[1])
+    self.assertEqual(308, message.repeated_fixed64[1])
+    self.assertEqual(309, message.repeated_sfixed32[1])
+    self.assertEqual(310, message.repeated_sfixed64[1])
+    self.assertEqual(311, message.repeated_float[1])
+    self.assertEqual(312, message.repeated_double[1])
+    self.assertEqual(False, message.repeated_bool[1])
+    self.assertEqual('315', message.repeated_string[1])
+    self.assertEqual('316', message.repeated_bytes[1])
+
+    self.assertEqual(317, message.repeatedgroup[1].a)
+    self.assertEqual(318, message.repeated_nested_message[1].bb)
+    self.assertEqual(319, message.repeated_foreign_message[1].c)
+    self.assertEqual(320, message.repeated_import_message[1].d)
+
+    self.assertEqual(unittest_pb2.TestAllTypes.BAZ,
+                     message.repeated_nested_enum[1])
+    self.assertEqual(unittest_pb2.FOREIGN_BAZ,
+                     message.repeated_foreign_enum[1])
+    self.assertEqual(unittest_import_pb2.IMPORT_BAZ,
+                     message.repeated_import_enum[1])
+
+    # -----------------------------------------------------------------
+
+    self.assertTrue(message.HasField('default_int32'))
+    self.assertTrue(message.HasField('default_int64'))
+    self.assertTrue(message.HasField('default_uint32'))
+    self.assertTrue(message.HasField('default_uint64'))
+    self.assertTrue(message.HasField('default_sint32'))
+    self.assertTrue(message.HasField('default_sint64'))
+    self.assertTrue(message.HasField('default_fixed32'))
+    self.assertTrue(message.HasField('default_fixed64'))
+    self.assertTrue(message.HasField('default_sfixed32'))
+    self.assertTrue(message.HasField('default_sfixed64'))
+    self.assertTrue(message.HasField('default_float'))
+    self.assertTrue(message.HasField('default_double'))
+    self.assertTrue(message.HasField('default_bool'))
+    self.assertTrue(message.HasField('default_string'))
+    self.assertTrue(message.HasField('default_bytes'))
+
+    self.assertTrue(message.HasField('default_nested_enum'))
+    self.assertTrue(message.HasField('default_foreign_enum'))
+    self.assertTrue(message.HasField('default_import_enum'))
+
+    self.assertEqual(401, message.default_int32)
+    self.assertEqual(402, message.default_int64)
+    self.assertEqual(403, message.default_uint32)
+    self.assertEqual(404, message.default_uint64)
+    self.assertEqual(405, message.default_sint32)
+    self.assertEqual(406, message.default_sint64)
+    self.assertEqual(407, message.default_fixed32)
+    self.assertEqual(408, message.default_fixed64)
+    self.assertEqual(409, message.default_sfixed32)
+    self.assertEqual(410, message.default_sfixed64)
+    self.assertEqual(411, message.default_float)
+    self.assertEqual(412, message.default_double)
+    self.assertEqual(False, message.default_bool)
+    self.assertEqual('415', message.default_string)
+    self.assertEqual('416', message.default_bytes)
+
+    self.assertEqual(unittest_pb2.TestAllTypes.FOO, message.default_nested_enum)
+    self.assertEqual(unittest_pb2.FOREIGN_FOO, message.default_foreign_enum)
+    self.assertEqual(unittest_import_pb2.IMPORT_FOO,
+                     message.default_import_enum)
+
+def GoldenFile(filename):
+  """Finds the given golden file and returns a file object representing it."""
+
+  # Search up the directory tree looking for the C++ protobuf source code.
+  path = '.'
+  while os.path.exists(path):
+    if os.path.exists(os.path.join(path, 'src/google/protobuf')):
+      # Found it.  Load the golden file from the testdata directory.
+      full_path = os.path.join(path, 'src/google/protobuf/testdata', filename)
+      return open(full_path, 'rb')
+    path = os.path.join(path, '..')
+
+  raise RuntimeError(
+    'Could not find golden files.  This test must be run from within the '
+    'protobuf source package so that it can read test data files from the '
+    'C++ source tree.')
+
+
+def SetAllPackedFields(message):
+  """Sets every field in the message to a unique value.
+
+  Args:
+    message: A unittest_pb2.TestPackedTypes instance.
+  """
+  message.packed_int32.extend([101, 102])
+  message.packed_int64.extend([103, 104])
+  message.packed_uint32.extend([105, 106])
+  message.packed_uint64.extend([107, 108])
+  message.packed_sint32.extend([109, 110])
+  message.packed_sint64.extend([111, 112])
+  message.packed_fixed32.extend([113, 114])
+  message.packed_fixed64.extend([115, 116])
+  message.packed_sfixed32.extend([117, 118])
+  message.packed_sfixed64.extend([119, 120])
+  message.packed_float.extend([121.0, 122.0])
+  message.packed_double.extend([122.0, 123.0])
+  message.packed_bool.extend([True, False])
+  message.packed_enum.extend([unittest_pb2.FOREIGN_FOO,
+                              unittest_pb2.FOREIGN_BAR])
+
+
+def SetAllPackedExtensions(message):
+  """Sets every extension in the message to a unique value.
+
+  Args:
+    message: A unittest_pb2.TestPackedExtensions instance.
+  """
+  extensions = message.Extensions
+  pb2 = unittest_pb2
+
+  extensions[pb2.packed_int32_extension].append(101)
+  extensions[pb2.packed_int64_extension].append(102)
+  extensions[pb2.packed_uint32_extension].append(103)
+  extensions[pb2.packed_uint64_extension].append(104)
+  extensions[pb2.packed_sint32_extension].append(105)
+  extensions[pb2.packed_sint64_extension].append(106)
+  extensions[pb2.packed_fixed32_extension].append(107)
+  extensions[pb2.packed_fixed64_extension].append(108)
+  extensions[pb2.packed_sfixed32_extension].append(109)
+  extensions[pb2.packed_sfixed64_extension].append(110)
+  extensions[pb2.packed_float_extension].append(111.0)
+  extensions[pb2.packed_double_extension].append(112.0)
+  extensions[pb2.packed_bool_extension].append(True)
+  extensions[pb2.packed_enum_extension].append(pb2.FOREIGN_BAZ)
diff --git a/python/google/protobuf/internal/text_format_test.py b/python/google/protobuf/internal/text_format_test.py
new file mode 100755
index 0000000..0cf2718
--- /dev/null
+++ b/python/google/protobuf/internal/text_format_test.py
@@ -0,0 +1,397 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.text_format."""
+
+__author__ = 'kenton@google.com (Kenton Varda)'
+
+import difflib
+
+import unittest
+from google.protobuf import text_format
+from google.protobuf.internal import test_util
+from google.protobuf import unittest_pb2
+from google.protobuf import unittest_mset_pb2
+
+
+class TextFormatTest(test_util.GoldenMessageTestCase):
+  def ReadGolden(self, golden_filename):
+    f = test_util.GoldenFile(golden_filename)
+    golden_lines = f.readlines()
+    f.close()
+    return golden_lines
+
+  def CompareToGoldenFile(self, text, golden_filename):
+    golden_lines = self.ReadGolden(golden_filename)
+    self.CompareToGoldenLines(text, golden_lines)
+
+  def CompareToGoldenText(self, text, golden_text):
+    self.CompareToGoldenLines(text, golden_text.splitlines(1))
+
+  def CompareToGoldenLines(self, text, golden_lines):
+    actual_lines = text.splitlines(1)
+    self.assertEqual(golden_lines, actual_lines,
+      "Text doesn't match golden.  Diff:\n" +
+      ''.join(difflib.ndiff(golden_lines, actual_lines)))
+
+  def testPrintAllFields(self):
+    message = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(message)
+    self.CompareToGoldenFile(
+      self.RemoveRedundantZeros(text_format.MessageToString(message)),
+      'text_format_unittest_data.txt')
+
+  def testPrintAllExtensions(self):
+    message = unittest_pb2.TestAllExtensions()
+    test_util.SetAllExtensions(message)
+    self.CompareToGoldenFile(
+      self.RemoveRedundantZeros(text_format.MessageToString(message)),
+      'text_format_unittest_extensions_data.txt')
+
+  def testPrintMessageSet(self):
+    message = unittest_mset_pb2.TestMessageSetContainer()
+    ext1 = unittest_mset_pb2.TestMessageSetExtension1.message_set_extension
+    ext2 = unittest_mset_pb2.TestMessageSetExtension2.message_set_extension
+    message.message_set.Extensions[ext1].i = 23
+    message.message_set.Extensions[ext2].str = 'foo'
+    self.CompareToGoldenText(text_format.MessageToString(message),
+      'message_set {\n'
+      '  [protobuf_unittest.TestMessageSetExtension1] {\n'
+      '    i: 23\n'
+      '  }\n'
+      '  [protobuf_unittest.TestMessageSetExtension2] {\n'
+      '    str: \"foo\"\n'
+      '  }\n'
+      '}\n')
+
+  def testPrintExotic(self):
+    message = unittest_pb2.TestAllTypes()
+    message.repeated_int64.append(-9223372036854775808);
+    message.repeated_uint64.append(18446744073709551615);
+    message.repeated_double.append(123.456);
+    message.repeated_double.append(1.23e22);
+    message.repeated_double.append(1.23e-18);
+    message.repeated_string.append('\000\001\a\b\f\n\r\t\v\\\'\"');
+    self.CompareToGoldenText(
+      self.RemoveRedundantZeros(text_format.MessageToString(message)),
+      'repeated_int64: -9223372036854775808\n'
+      'repeated_uint64: 18446744073709551615\n'
+      'repeated_double: 123.456\n'
+      'repeated_double: 1.23e+22\n'
+      'repeated_double: 1.23e-18\n'
+      'repeated_string: '
+        '\"\\000\\001\\007\\010\\014\\n\\r\\t\\013\\\\\\\'\\\"\"\n')
+
+  def testMessageToString(self):
+    message = unittest_pb2.ForeignMessage()
+    message.c = 123
+    self.assertEqual('c: 123\n', str(message))
+
+  def RemoveRedundantZeros(self, text):
+    # Some platforms print 1e+5 as 1e+005.  This is fine, but we need to remove
+    # these zeros in order to match the golden file.
+    return text.replace('e+0','e+').replace('e+0','e+') \
+               .replace('e-0','e-').replace('e-0','e-')
+
+  def testMergeGolden(self):
+    golden_text = '\n'.join(self.ReadGolden('text_format_unittest_data.txt'))
+    parsed_message = unittest_pb2.TestAllTypes()
+    text_format.Merge(golden_text, parsed_message)
+
+    message = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(message)
+    self.assertEquals(message, parsed_message)
+
+  def testMergeGoldenExtensions(self):
+    golden_text = '\n'.join(self.ReadGolden(
+        'text_format_unittest_extensions_data.txt'))
+    parsed_message = unittest_pb2.TestAllExtensions()
+    text_format.Merge(golden_text, parsed_message)
+
+    message = unittest_pb2.TestAllExtensions()
+    test_util.SetAllExtensions(message)
+    self.assertEquals(message, parsed_message)
+
+  def testMergeAllFields(self):
+    message = unittest_pb2.TestAllTypes()
+    test_util.SetAllFields(message)
+    ascii_text = text_format.MessageToString(message)
+
+    parsed_message = unittest_pb2.TestAllTypes()
+    text_format.Merge(ascii_text, parsed_message)
+    self.assertEqual(message, parsed_message)
+    self.ExpectAllFieldsSet(message)
+
+  def testMergeAllExtensions(self):
+    message = unittest_pb2.TestAllExtensions()
+    test_util.SetAllExtensions(message)
+    ascii_text = text_format.MessageToString(message)
+
+    parsed_message = unittest_pb2.TestAllExtensions()
+    text_format.Merge(ascii_text, parsed_message)
+    self.assertEqual(message, parsed_message)
+
+  def testMergeMessageSet(self):
+    message = unittest_pb2.TestAllTypes()
+    text = ('repeated_uint64: 1\n'
+            'repeated_uint64: 2\n')
+    text_format.Merge(text, message)
+    self.assertEqual(1, message.repeated_uint64[0])
+    self.assertEqual(2, message.repeated_uint64[1])
+
+    message = unittest_mset_pb2.TestMessageSetContainer()
+    text = ('message_set {\n'
+            '  [protobuf_unittest.TestMessageSetExtension1] {\n'
+            '    i: 23\n'
+            '  }\n'
+            '  [protobuf_unittest.TestMessageSetExtension2] {\n'
+            '    str: \"foo\"\n'
+            '  }\n'
+            '}\n')
+    text_format.Merge(text, message)
+    ext1 = unittest_mset_pb2.TestMessageSetExtension1.message_set_extension
+    ext2 = unittest_mset_pb2.TestMessageSetExtension2.message_set_extension
+    self.assertEquals(23, message.message_set.Extensions[ext1].i)
+    self.assertEquals('foo', message.message_set.Extensions[ext2].str)
+
+  def testMergeExotic(self):
+    message = unittest_pb2.TestAllTypes()
+    text = ('repeated_int64: -9223372036854775808\n'
+            'repeated_uint64: 18446744073709551615\n'
+            'repeated_double: 123.456\n'
+            'repeated_double: 1.23e+22\n'
+            'repeated_double: 1.23e-18\n'
+            'repeated_string: \n'
+            '\"\\000\\001\\007\\010\\014\\n\\r\\t\\013\\\\\\\'\\\"\"\n')
+    text_format.Merge(text, message)
+
+    self.assertEqual(-9223372036854775808, message.repeated_int64[0])
+    self.assertEqual(18446744073709551615, message.repeated_uint64[0])
+    self.assertEqual(123.456, message.repeated_double[0])
+    self.assertEqual(1.23e22, message.repeated_double[1])
+    self.assertEqual(1.23e-18, message.repeated_double[2])
+    self.assertEqual(
+        '\000\001\a\b\f\n\r\t\v\\\'\"', message.repeated_string[0])
+
+  def testMergeUnknownField(self):
+    message = unittest_pb2.TestAllTypes()
+    text = 'unknown_field: 8\n'
+    self.assertRaisesWithMessage(
+        text_format.ParseError,
+        ('1:1 : Message type "protobuf_unittest.TestAllTypes" has no field named '
+         '"unknown_field".'),
+        text_format.Merge, text, message)
+
+  def testMergeBadExtension(self):
+    message = unittest_pb2.TestAllTypes()
+    text = '[unknown_extension]: 8\n'
+    self.assertRaisesWithMessage(
+        text_format.ParseError,
+        '1:2 : Extension "unknown_extension" not registered.',
+        text_format.Merge, text, message)
+
+  def testMergeGroupNotClosed(self):
+    message = unittest_pb2.TestAllTypes()
+    text = 'RepeatedGroup: <'
+    self.assertRaisesWithMessage(
+        text_format.ParseError, '1:16 : Expected ">".',
+        text_format.Merge, text, message)
+
+    text = 'RepeatedGroup: {'
+    self.assertRaisesWithMessage(
+        text_format.ParseError, '1:16 : Expected "}".',
+        text_format.Merge, text, message)
+
+  def testMergeBadEnumValue(self):
+    message = unittest_pb2.TestAllTypes()
+    text = 'optional_nested_enum: BARR'
+    self.assertRaisesWithMessage(
+        text_format.ParseError,
+        ('1:23 : Enum type "protobuf_unittest.TestAllTypes.NestedEnum" '
+         'has no value named BARR.'),
+        text_format.Merge, text, message)
+
+    message = unittest_pb2.TestAllTypes()
+    text = 'optional_nested_enum: 100'
+    self.assertRaisesWithMessage(
+        text_format.ParseError,
+        ('1:23 : Enum type "protobuf_unittest.TestAllTypes.NestedEnum" '
+         'has no value with number 100.'),
+        text_format.Merge, text, message)
+
+  def assertRaisesWithMessage(self, e_class, e, func, *args, **kwargs):
+    """Same as assertRaises, but also compares the exception message."""
+    if hasattr(e_class, '__name__'):
+      exc_name = e_class.__name__
+    else:
+      exc_name = str(e_class)
+
+    try:
+      func(*args, **kwargs)
+    except e_class, expr:
+      if str(expr) != e:
+        msg = '%s raised, but with wrong message: "%s" instead of "%s"'
+        raise self.failureException(msg % (exc_name,
+                                           str(expr).encode('string_escape'),
+                                           e.encode('string_escape')))
+      return
+    else:
+      raise self.failureException('%s not raised' % exc_name)
+
+
+class TokenizerTest(unittest.TestCase):
+
+  def testSimpleTokenCases(self):
+    text = ('identifier1:"string1"\n     \n\n'
+            'identifier2 : \n \n123  \n  identifier3 :\'string\'\n'
+            'identifiER_4 : 1.1e+2 ID5:-0.23 ID6:\'aaaa\\\'bbbb\'\n'
+            'ID7 : "aa\\"bb"\n\n\n\n ID8: {A:inf B:-inf C:true D:false}\n'
+            'ID9: 22 ID10: -111111111111111111 ID11: -22\n'
+            'ID12: 2222222222222222222')
+    tokenizer = text_format._Tokenizer(text)
+    methods = [(tokenizer.ConsumeIdentifier, 'identifier1'),
+               ':',
+               (tokenizer.ConsumeString, 'string1'),
+               (tokenizer.ConsumeIdentifier, 'identifier2'),
+               ':',
+               (tokenizer.ConsumeInt32, 123),
+               (tokenizer.ConsumeIdentifier, 'identifier3'),
+               ':',
+               (tokenizer.ConsumeString, 'string'),
+               (tokenizer.ConsumeIdentifier, 'identifiER_4'),
+               ':',
+               (tokenizer.ConsumeFloat, 1.1e+2),
+               (tokenizer.ConsumeIdentifier, 'ID5'),
+               ':',
+               (tokenizer.ConsumeFloat, -0.23),
+               (tokenizer.ConsumeIdentifier, 'ID6'),
+               ':',
+               (tokenizer.ConsumeString, 'aaaa\'bbbb'),
+               (tokenizer.ConsumeIdentifier, 'ID7'),
+               ':',
+               (tokenizer.ConsumeString, 'aa\"bb'),
+               (tokenizer.ConsumeIdentifier, 'ID8'),
+               ':',
+               '{',
+               (tokenizer.ConsumeIdentifier, 'A'),
+               ':',
+               (tokenizer.ConsumeFloat, float('inf')),
+               (tokenizer.ConsumeIdentifier, 'B'),
+               ':',
+               (tokenizer.ConsumeFloat, float('-inf')),
+               (tokenizer.ConsumeIdentifier, 'C'),
+               ':',
+               (tokenizer.ConsumeBool, True),
+               (tokenizer.ConsumeIdentifier, 'D'),
+               ':',
+               (tokenizer.ConsumeBool, False),
+               '}',
+               (tokenizer.ConsumeIdentifier, 'ID9'),
+               ':',
+               (tokenizer.ConsumeUint32, 22),
+               (tokenizer.ConsumeIdentifier, 'ID10'),
+               ':',
+               (tokenizer.ConsumeInt64, -111111111111111111),
+               (tokenizer.ConsumeIdentifier, 'ID11'),
+               ':',
+               (tokenizer.ConsumeInt32, -22),
+               (tokenizer.ConsumeIdentifier, 'ID12'),
+               ':',
+               (tokenizer.ConsumeUint64, 2222222222222222222)]
+
+    i = 0
+    while not tokenizer.AtEnd():
+      m = methods[i]
+      if type(m) == str:
+        token = tokenizer.token
+        self.assertEqual(token, m)
+        tokenizer.NextToken()
+      else:
+        self.assertEqual(m[1], m[0]())
+      i += 1
+
+  def testConsumeIntegers(self):
+    # This test only tests the failures in the integer parsing methods as well
+    # as the '0' special cases.
+    int64_max = (1 << 63) - 1
+    uint32_max = (1 << 32) - 1
+    text = '-1 %d %d' % (uint32_max + 1, int64_max + 1)
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeUint32)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeUint64)
+    self.assertEqual(-1, tokenizer.ConsumeInt32())
+
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeUint32)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeInt32)
+    self.assertEqual(uint32_max + 1, tokenizer.ConsumeInt64())
+
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeInt64)
+    self.assertEqual(int64_max + 1, tokenizer.ConsumeUint64())
+    self.assertTrue(tokenizer.AtEnd())
+
+    text = '-0 -0 0 0'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertEqual(0, tokenizer.ConsumeUint32())
+    self.assertEqual(0, tokenizer.ConsumeUint64())
+    self.assertEqual(0, tokenizer.ConsumeUint32())
+    self.assertEqual(0, tokenizer.ConsumeUint64())
+    self.assertTrue(tokenizer.AtEnd())
+
+  def testConsumeByteString(self):
+    text = '"string1\''
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeByteString)
+
+    text = 'string1"'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeByteString)
+
+    text = '\n"\\xt"'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeByteString)
+
+    text = '\n"\\"'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeByteString)
+
+    text = '\n"\\x"'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeByteString)
+
+  def testConsumeBool(self):
+    text = 'not-a-bool'
+    tokenizer = text_format._Tokenizer(text)
+    self.assertRaises(text_format.ParseError, tokenizer.ConsumeBool)
+
+
+if __name__ == '__main__':
+  unittest.main()
diff --git a/python/google/protobuf/internal/type_checkers.py b/python/google/protobuf/internal/type_checkers.py
new file mode 100755
index 0000000..a3bc57f
--- /dev/null
+++ b/python/google/protobuf/internal/type_checkers.py
@@ -0,0 +1,287 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Provides type checking routines.
+
+This module defines type checking utilities in the forms of dictionaries:
+
+VALUE_CHECKERS: A dictionary of field types and a value validation object.
+TYPE_TO_BYTE_SIZE_FN: A dictionary with field types and a size computing
+  function.
+TYPE_TO_SERIALIZE_METHOD: A dictionary with field types and serialization
+  function.
+FIELD_TYPE_TO_WIRE_TYPE: A dictionary with field typed and their
+  coresponding wire types.
+TYPE_TO_DESERIALIZE_METHOD: A dictionary with field types and deserialization
+  function.
+"""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+from google.protobuf.internal import decoder
+from google.protobuf.internal import encoder
+from google.protobuf.internal import wire_format
+from google.protobuf import descriptor
+
+_FieldDescriptor = descriptor.FieldDescriptor
+
+
+def GetTypeChecker(cpp_type, field_type):
+  """Returns a type checker for a message field of the specified types.
+
+  Args:
+    cpp_type: C++ type of the field (see descriptor.py).
+    field_type: Protocol message field type (see descriptor.py).
+
+  Returns:
+    An instance of TypeChecker which can be used to verify the types
+    of values assigned to a field of the specified type.
+  """
+  if (cpp_type == _FieldDescriptor.CPPTYPE_STRING and
+      field_type == _FieldDescriptor.TYPE_STRING):
+    return UnicodeValueChecker()
+  return _VALUE_CHECKERS[cpp_type]
+
+
+# None of the typecheckers below make any attempt to guard against people
+# subclassing builtin types and doing weird things.  We're not trying to
+# protect against malicious clients here, just people accidentally shooting
+# themselves in the foot in obvious ways.
+
+class TypeChecker(object):
+
+  """Type checker used to catch type errors as early as possible
+  when the client is setting scalar fields in protocol messages.
+  """
+
+  def __init__(self, *acceptable_types):
+    self._acceptable_types = acceptable_types
+
+  def CheckValue(self, proposed_value):
+    if not isinstance(proposed_value, self._acceptable_types):
+      message = ('%.1024r has type %s, but expected one of: %s' %
+                 (proposed_value, type(proposed_value), self._acceptable_types))
+      raise TypeError(message)
+
+
+# IntValueChecker and its subclasses perform integer type-checks
+# and bounds-checks.
+class IntValueChecker(object):
+
+  """Checker used for integer fields.  Performs type-check and range check."""
+
+  def CheckValue(self, proposed_value):
+    if not isinstance(proposed_value, (int, long)):
+      message = ('%.1024r has type %s, but expected one of: %s' %
+                 (proposed_value, type(proposed_value), (int, long)))
+      raise TypeError(message)
+    if not self._MIN <= proposed_value <= self._MAX:
+      raise ValueError('Value out of range: %d' % proposed_value)
+
+
+class UnicodeValueChecker(object):
+
+  """Checker used for string fields."""
+
+  def CheckValue(self, proposed_value):
+    if not isinstance(proposed_value, (str, unicode)):
+      message = ('%.1024r has type %s, but expected one of: %s' %
+                 (proposed_value, type(proposed_value), (str, unicode)))
+      raise TypeError(message)
+
+    # If the value is of type 'str' make sure that it is in 7-bit ASCII
+    # encoding.
+    if isinstance(proposed_value, str):
+      try:
+        unicode(proposed_value, 'ascii')
+      except UnicodeDecodeError:
+        raise ValueError('%.1024r has type str, but isn\'t in 7-bit ASCII '
+                         'encoding. Non-ASCII strings must be converted to '
+                         'unicode objects before being added.' %
+                         (proposed_value))
+
+
+class Int32ValueChecker(IntValueChecker):
+  # We're sure to use ints instead of longs here since comparison may be more
+  # efficient.
+  _MIN = -2147483648
+  _MAX = 2147483647
+
+
+class Uint32ValueChecker(IntValueChecker):
+  _MIN = 0
+  _MAX = (1 << 32) - 1
+
+
+class Int64ValueChecker(IntValueChecker):
+  _MIN = -(1 << 63)
+  _MAX = (1 << 63) - 1
+
+
+class Uint64ValueChecker(IntValueChecker):
+  _MIN = 0
+  _MAX = (1 << 64) - 1
+
+
+# Type-checkers for all scalar CPPTYPEs.
+_VALUE_CHECKERS = {
+    _FieldDescriptor.CPPTYPE_INT32: Int32ValueChecker(),
+    _FieldDescriptor.CPPTYPE_INT64: Int64ValueChecker(),
+    _FieldDescriptor.CPPTYPE_UINT32: Uint32ValueChecker(),
+    _FieldDescriptor.CPPTYPE_UINT64: Uint64ValueChecker(),
+    _FieldDescriptor.CPPTYPE_DOUBLE: TypeChecker(
+        float, int, long),
+    _FieldDescriptor.CPPTYPE_FLOAT: TypeChecker(
+        float, int, long),
+    _FieldDescriptor.CPPTYPE_BOOL: TypeChecker(bool, int),
+    _FieldDescriptor.CPPTYPE_ENUM: Int32ValueChecker(),
+    _FieldDescriptor.CPPTYPE_STRING: TypeChecker(str),
+    }
+
+
+# Map from field type to a function F, such that F(field_num, value)
+# gives the total byte size for a value of the given type.  This
+# byte size includes tag information and any other additional space
+# associated with serializing "value".
+TYPE_TO_BYTE_SIZE_FN = {
+    _FieldDescriptor.TYPE_DOUBLE: wire_format.DoubleByteSize,
+    _FieldDescriptor.TYPE_FLOAT: wire_format.FloatByteSize,
+    _FieldDescriptor.TYPE_INT64: wire_format.Int64ByteSize,
+    _FieldDescriptor.TYPE_UINT64: wire_format.UInt64ByteSize,
+    _FieldDescriptor.TYPE_INT32: wire_format.Int32ByteSize,
+    _FieldDescriptor.TYPE_FIXED64: wire_format.Fixed64ByteSize,
+    _FieldDescriptor.TYPE_FIXED32: wire_format.Fixed32ByteSize,
+    _FieldDescriptor.TYPE_BOOL: wire_format.BoolByteSize,
+    _FieldDescriptor.TYPE_STRING: wire_format.StringByteSize,
+    _FieldDescriptor.TYPE_GROUP: wire_format.GroupByteSize,
+    _FieldDescriptor.TYPE_MESSAGE: wire_format.MessageByteSize,
+    _FieldDescriptor.TYPE_BYTES: wire_format.BytesByteSize,
+    _FieldDescriptor.TYPE_UINT32: wire_format.UInt32ByteSize,
+    _FieldDescriptor.TYPE_ENUM: wire_format.EnumByteSize,
+    _FieldDescriptor.TYPE_SFIXED32: wire_format.SFixed32ByteSize,
+    _FieldDescriptor.TYPE_SFIXED64: wire_format.SFixed64ByteSize,
+    _FieldDescriptor.TYPE_SINT32: wire_format.SInt32ByteSize,
+    _FieldDescriptor.TYPE_SINT64: wire_format.SInt64ByteSize
+    }
+
+
+# Maps from field type to an unbound Encoder method F, such that
+# F(encoder, field_number, value) will append the serialization
+# of a value of this type to the encoder.
+_Encoder = encoder.Encoder
+TYPE_TO_SERIALIZE_METHOD = {
+    _FieldDescriptor.TYPE_DOUBLE: _Encoder.AppendDouble,
+    _FieldDescriptor.TYPE_FLOAT: _Encoder.AppendFloat,
+    _FieldDescriptor.TYPE_INT64: _Encoder.AppendInt64,
+    _FieldDescriptor.TYPE_UINT64: _Encoder.AppendUInt64,
+    _FieldDescriptor.TYPE_INT32: _Encoder.AppendInt32,
+    _FieldDescriptor.TYPE_FIXED64: _Encoder.AppendFixed64,
+    _FieldDescriptor.TYPE_FIXED32: _Encoder.AppendFixed32,
+    _FieldDescriptor.TYPE_BOOL: _Encoder.AppendBool,
+    _FieldDescriptor.TYPE_STRING: _Encoder.AppendString,
+    _FieldDescriptor.TYPE_GROUP: _Encoder.AppendGroup,
+    _FieldDescriptor.TYPE_MESSAGE: _Encoder.AppendMessage,
+    _FieldDescriptor.TYPE_BYTES: _Encoder.AppendBytes,
+    _FieldDescriptor.TYPE_UINT32: _Encoder.AppendUInt32,
+    _FieldDescriptor.TYPE_ENUM: _Encoder.AppendEnum,
+    _FieldDescriptor.TYPE_SFIXED32: _Encoder.AppendSFixed32,
+    _FieldDescriptor.TYPE_SFIXED64: _Encoder.AppendSFixed64,
+    _FieldDescriptor.TYPE_SINT32: _Encoder.AppendSInt32,
+    _FieldDescriptor.TYPE_SINT64: _Encoder.AppendSInt64,
+    }
+
+
+TYPE_TO_NOTAG_SERIALIZE_METHOD = {
+    _FieldDescriptor.TYPE_DOUBLE: _Encoder.AppendDoubleNoTag,
+    _FieldDescriptor.TYPE_FLOAT: _Encoder.AppendFloatNoTag,
+    _FieldDescriptor.TYPE_INT64: _Encoder.AppendInt64NoTag,
+    _FieldDescriptor.TYPE_UINT64: _Encoder.AppendUInt64NoTag,
+    _FieldDescriptor.TYPE_INT32: _Encoder.AppendInt32NoTag,
+    _FieldDescriptor.TYPE_FIXED64: _Encoder.AppendFixed64NoTag,
+    _FieldDescriptor.TYPE_FIXED32: _Encoder.AppendFixed32NoTag,
+    _FieldDescriptor.TYPE_BOOL: _Encoder.AppendBoolNoTag,
+    _FieldDescriptor.TYPE_UINT32: _Encoder.AppendUInt32NoTag,
+    _FieldDescriptor.TYPE_ENUM: _Encoder.AppendEnumNoTag,
+    _FieldDescriptor.TYPE_SFIXED32: _Encoder.AppendSFixed32NoTag,
+    _FieldDescriptor.TYPE_SFIXED64: _Encoder.AppendSFixed64NoTag,
+    _FieldDescriptor.TYPE_SINT32: _Encoder.AppendSInt32NoTag,
+    _FieldDescriptor.TYPE_SINT64: _Encoder.AppendSInt64NoTag,
+    }
+
+# Maps from field type to expected wiretype.
+FIELD_TYPE_TO_WIRE_TYPE = {
+    _FieldDescriptor.TYPE_DOUBLE: wire_format.WIRETYPE_FIXED64,
+    _FieldDescriptor.TYPE_FLOAT: wire_format.WIRETYPE_FIXED32,
+    _FieldDescriptor.TYPE_INT64: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_UINT64: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_INT32: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_FIXED64: wire_format.WIRETYPE_FIXED64,
+    _FieldDescriptor.TYPE_FIXED32: wire_format.WIRETYPE_FIXED32,
+    _FieldDescriptor.TYPE_BOOL: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_STRING:
+      wire_format.WIRETYPE_LENGTH_DELIMITED,
+    _FieldDescriptor.TYPE_GROUP: wire_format.WIRETYPE_START_GROUP,
+    _FieldDescriptor.TYPE_MESSAGE:
+      wire_format.WIRETYPE_LENGTH_DELIMITED,
+    _FieldDescriptor.TYPE_BYTES:
+      wire_format.WIRETYPE_LENGTH_DELIMITED,
+    _FieldDescriptor.TYPE_UINT32: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_ENUM: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_SFIXED32: wire_format.WIRETYPE_FIXED32,
+    _FieldDescriptor.TYPE_SFIXED64: wire_format.WIRETYPE_FIXED64,
+    _FieldDescriptor.TYPE_SINT32: wire_format.WIRETYPE_VARINT,
+    _FieldDescriptor.TYPE_SINT64: wire_format.WIRETYPE_VARINT,
+    }
+
+
+# Maps from field type to an unbound Decoder method F,
+# such that F(decoder) will read a field of the requested type.
+#
+# Note that Message and Group are intentionally missing here.
+# They're handled by _RecursivelyMerge().
+_Decoder = decoder.Decoder
+TYPE_TO_DESERIALIZE_METHOD = {
+    _FieldDescriptor.TYPE_DOUBLE: _Decoder.ReadDouble,
+    _FieldDescriptor.TYPE_FLOAT: _Decoder.ReadFloat,
+    _FieldDescriptor.TYPE_INT64: _Decoder.ReadInt64,
+    _FieldDescriptor.TYPE_UINT64: _Decoder.ReadUInt64,
+    _FieldDescriptor.TYPE_INT32: _Decoder.ReadInt32,
+    _FieldDescriptor.TYPE_FIXED64: _Decoder.ReadFixed64,
+    _FieldDescriptor.TYPE_FIXED32: _Decoder.ReadFixed32,
+    _FieldDescriptor.TYPE_BOOL: _Decoder.ReadBool,
+    _FieldDescriptor.TYPE_STRING: _Decoder.ReadString,
+    _FieldDescriptor.TYPE_BYTES: _Decoder.ReadBytes,
+    _FieldDescriptor.TYPE_UINT32: _Decoder.ReadUInt32,
+    _FieldDescriptor.TYPE_ENUM: _Decoder.ReadEnum,
+    _FieldDescriptor.TYPE_SFIXED32: _Decoder.ReadSFixed32,
+    _FieldDescriptor.TYPE_SFIXED64: _Decoder.ReadSFixed64,
+    _FieldDescriptor.TYPE_SINT32: _Decoder.ReadSInt32,
+    _FieldDescriptor.TYPE_SINT64: _Decoder.ReadSInt64,
+    }
diff --git a/python/google/protobuf/internal/wire_format.py b/python/google/protobuf/internal/wire_format.py
new file mode 100755
index 0000000..da6464d
--- /dev/null
+++ b/python/google/protobuf/internal/wire_format.py
@@ -0,0 +1,247 @@
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Constants and static functions to support protocol buffer wire format."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import struct
+from google.protobuf import message
+
+
+TAG_TYPE_BITS = 3  # Number of bits used to hold type info in a proto tag.
+_TAG_TYPE_MASK = (1 << TAG_TYPE_BITS) - 1  # 0x7
+
+# These numbers identify the wire type of a protocol buffer value.
+# We use the least-significant TAG_TYPE_BITS bits of the varint-encoded
+# tag-and-type to store one of these WIRETYPE_* constants.
+# These values must match WireType enum in //net/proto2/public/wire_format.h.
+WIRETYPE_VARINT = 0
+WIRETYPE_FIXED64 = 1
+WIRETYPE_LENGTH_DELIMITED = 2
+WIRETYPE_START_GROUP = 3
+WIRETYPE_END_GROUP = 4
+WIRETYPE_FIXED32 = 5
+_WIRETYPE_MAX = 5
+
+
+# Bounds for various integer types.
+INT32_MAX = int((1 << 31) - 1)
+INT32_MIN = int(-(1 << 31))
+UINT32_MAX = (1 << 32) - 1
+
+INT64_MAX = (1 << 63) - 1
+INT64_MIN = -(1 << 63)
+UINT64_MAX = (1 << 64) - 1
+
+# "struct" format strings that will encode/decode the specified formats.
+FORMAT_UINT32_LITTLE_ENDIAN = '<I'
+FORMAT_UINT64_LITTLE_ENDIAN = '<Q'
+FORMAT_FLOAT_LITTLE_ENDIAN = '<f'
+FORMAT_DOUBLE_LITTLE_ENDIAN = '<d'
+
+
+# We'll have to provide alternate implementations of AppendLittleEndian*() on
+# any architectures where these checks fail.
+if struct.calcsize(FORMAT_UINT32_LITTLE_ENDIAN) != 4:
+  raise AssertionError('Format "I" is not a 32-bit number.')
+if struct.calcsize(FORMAT_UINT64_LITTLE_ENDIAN) != 8:
+  raise AssertionError('Format "Q" is not a 64-bit number.')
+
+
+def PackTag(field_number, wire_type):
+  """Returns an unsigned 32-bit integer that encodes the field number and
+  wire type information in standard protocol message wire format.
+
+  Args:
+    field_number: Expected to be an integer in the range [1, 1 << 29)
+    wire_type: One of the WIRETYPE_* constants.
+  """
+  if not 0 <= wire_type <= _WIRETYPE_MAX:
+    raise message.EncodeError('Unknown wire type: %d' % wire_type)
+  return (field_number << TAG_TYPE_BITS) | wire_type
+
+
+def UnpackTag(tag):
+  """The inverse of PackTag().  Given an unsigned 32-bit number,
+  returns a (field_number, wire_type) tuple.
+  """
+  return (tag >> TAG_TYPE_BITS), (tag & _TAG_TYPE_MASK)
+
+
+def ZigZagEncode(value):
+  """ZigZag Transform:  Encodes signed integers so that they can be
+  effectively used with varint encoding.  See wire_format.h for
+  more details.
+  """
+  if value >= 0:
+    return value << 1
+  return (value << 1) ^ (~0)
+
+
+def ZigZagDecode(value):
+  """Inverse of ZigZagEncode()."""
+  if not value & 0x1:
+    return value >> 1
+  return (value >> 1) ^ (~0)
+
+
+
+# The *ByteSize() functions below return the number of bytes required to
+# serialize "field number + type" information and then serialize the value.
+
+
+def Int32ByteSize(field_number, int32):
+  return Int64ByteSize(field_number, int32)
+
+
+def Int32ByteSizeNoTag(int32):
+  return _VarUInt64ByteSizeNoTag(0xffffffffffffffff & int32)
+
+
+def Int64ByteSize(field_number, int64):
+  # Have to convert to uint before calling UInt64ByteSize().
+  return UInt64ByteSize(field_number, 0xffffffffffffffff & int64)
+
+
+def UInt32ByteSize(field_number, uint32):
+  return UInt64ByteSize(field_number, uint32)
+
+
+def UInt64ByteSize(field_number, uint64):
+  return TagByteSize(field_number) + _VarUInt64ByteSizeNoTag(uint64)
+
+
+def SInt32ByteSize(field_number, int32):
+  return UInt32ByteSize(field_number, ZigZagEncode(int32))
+
+
+def SInt64ByteSize(field_number, int64):
+  return UInt64ByteSize(field_number, ZigZagEncode(int64))
+
+
+def Fixed32ByteSize(field_number, fixed32):
+  return TagByteSize(field_number) + 4
+
+
+def Fixed64ByteSize(field_number, fixed64):
+  return TagByteSize(field_number) + 8
+
+
+def SFixed32ByteSize(field_number, sfixed32):
+  return TagByteSize(field_number) + 4
+
+
+def SFixed64ByteSize(field_number, sfixed64):
+  return TagByteSize(field_number) + 8
+
+
+def FloatByteSize(field_number, flt):
+  return TagByteSize(field_number) + 4
+
+
+def DoubleByteSize(field_number, double):
+  return TagByteSize(field_number) + 8
+
+
+def BoolByteSize(field_number, b):
+  return TagByteSize(field_number) + 1
+
+
+def EnumByteSize(field_number, enum):
+  return UInt32ByteSize(field_number, enum)
+
+
+def StringByteSize(field_number, string):
+  return BytesByteSize(field_number, string.encode('utf-8'))
+
+
+def BytesByteSize(field_number, b):
+  return (TagByteSize(field_number)
+          + _VarUInt64ByteSizeNoTag(len(b))
+          + len(b))
+
+
+def GroupByteSize(field_number, message):
+  return (2 * TagByteSize(field_number)  # START and END group.
+          + message.ByteSize())
+
+
+def MessageByteSize(field_number, message):
+  return (TagByteSize(field_number)
+          + _VarUInt64ByteSizeNoTag(message.ByteSize())
+          + message.ByteSize())
+
+
+def MessageSetItemByteSize(field_number, msg):
+  # First compute the sizes of the tags.
+  # There are 2 tags for the beginning and ending of the repeated group, that
+  # is field number 1, one with field number 2 (type_id) and one with field
+  # number 3 (message).
+  total_size = (2 * TagByteSize(1) + TagByteSize(2) + TagByteSize(3))
+
+  # Add the number of bytes for type_id.
+  total_size += _VarUInt64ByteSizeNoTag(field_number)
+
+  message_size = msg.ByteSize()
+
+  # The number of bytes for encoding the length of the message.
+  total_size += _VarUInt64ByteSizeNoTag(message_size)
+
+  # The size of the message.
+  total_size += message_size
+  return total_size
+
+
+def TagByteSize(field_number):
+  """Returns the bytes required to serialize a tag with this field number."""
+  # Just pass in type 0, since the type won't affect the tag+type size.
+  return _VarUInt64ByteSizeNoTag(PackTag(field_number, 0))
+
+
+# Private helper function for the *ByteSize() functions above.
+
+def _VarUInt64ByteSizeNoTag(uint64):
+  """Returns the number of bytes required to serialize a single varint
+  using boundary value comparisons. (unrolled loop optimization -WPierce)
+  uint64 must be unsigned.
+  """
+  if uint64 <= 0x7f: return 1
+  if uint64 <= 0x3fff: return 2
+  if uint64 <= 0x1fffff: return 3
+  if uint64 <= 0xfffffff: return 4
+  if uint64 <= 0x7ffffffff: return 5
+  if uint64 <= 0x3ffffffffff: return 6
+  if uint64 <= 0x1ffffffffffff: return 7
+  if uint64 <= 0xffffffffffffff: return 8
+  if uint64 <= 0x7fffffffffffffff: return 9
+  if uint64 > UINT64_MAX:
+    raise message.EncodeError('Value out of range: %d' % uint64)
+  return 10
diff --git a/python/google/protobuf/internal/wire_format_test.py b/python/google/protobuf/internal/wire_format_test.py
new file mode 100755
index 0000000..7600778
--- /dev/null
+++ b/python/google/protobuf/internal/wire_format_test.py
@@ -0,0 +1,253 @@
+#! /usr/bin/python
+#
+# Protocol Buffers - Google's data interchange format
+# Copyright 2008 Google Inc.  All rights reserved.
+# http://code.google.com/p/protobuf/
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+"""Test for google.protobuf.internal.wire_format."""
+
+__author__ = 'robinson@google.com (Will Robinson)'
+
+import unittest
+from google.protobuf import message
+from google.protobuf.internal import wire_format
+
+
+class WireFormatTest(unittest.TestCase):
+
+  def testPackTag(self):
+    field_number = 0xabc
+    tag_type = 2
+    self.assertEqual((field_number << 3) | tag_type,
+                     wire_format.PackTag(field_number, tag_type))
+    PackTag = wire_format.PackTag
+    # Number too high.
+    self.assertRaises(message.EncodeError, PackTag, field_number, 6)
+    # Number too low.
+    self.assertRaises(message.EncodeError, PackTag, field_number, -1)
+
+  def testUnpackTag(self):
+    # Test field numbers that will require various varint sizes.
+    for expected_field_number in (1, 15, 16, 2047, 2048):
+      for expected_wire_type in range(6):  # Highest-numbered wiretype is 5.
+        field_number, wire_type = wire_format.UnpackTag(
+            wire_format.PackTag(expected_field_number, expected_wire_type))
+        self.assertEqual(expected_field_number, field_number)
+        self.assertEqual(expected_wire_type, wire_type)
+
+    self.assertRaises(TypeError, wire_format.UnpackTag, None)
+    self.assertRaises(TypeError, wire_format.UnpackTag, 'abc')
+    self.assertRaises(TypeError, wire_format.UnpackTag, 0.0)
+    self.assertRaises(TypeError, wire_format.UnpackTag, object())
+
+  def testZigZagEncode(self):
+    Z = wire_format.ZigZagEncode
+    self.assertEqual(0, Z(0))
+    self.assertEqual(1, Z(-1))
+    self.assertEqual(2, Z(1))
+    self.assertEqual(3, Z(-2))
+    self.assertEqual(4, Z(2))
+    self.assertEqual(0xfffffffe, Z(0x7fffffff))
+    self.assertEqual(0xffffffff, Z(-0x80000000))
+    self.assertEqual(0xfffffffffffffffe, Z(0x7fffffffffffffff))
+    self.assertEqual(0xffffffffffffffff, Z(-0x8000000000000000))
+
+    self.assertRaises(TypeError, Z, None)
+    self.assertRaises(TypeError, Z, 'abcd')
+    self.assertRaises(TypeError, Z, 0.0)
+    self.assertRaises(TypeError, Z, object())
+
+  def testZigZagDecode(self):
+    Z = wire_format.ZigZagDecode
+    self.assertEqual(0, Z(0))
+    self.assertEqual(-1, Z(1))
+    self.assertEqual(1, Z(2))
+    self.assertEqual(-2, Z(3))
+    self.assertEqual(2, Z(4))
+    self.assertEqual(0x7fffffff, Z(0xfffffffe))
+    self.assertEqual(-0x80000000, Z(0xffffffff))
+    self.assertEqual(0x7fffffffffffffff, Z(0xfffffffffffffffe))
+    self.assertEqual(-0x8000000000000000, Z(0xffffffffffffffff))
+
+    self.assertRaises(TypeError, Z, None)
+    self.assertRaises(TypeError, Z, 'abcd')
+    self.assertRaises(TypeError, Z, 0.0)
+    self.assertRaises(TypeError, Z, object())
+
+  def NumericByteSizeTestHelper(self, byte_size_fn, value, expected_value_size):
+    # Use field numbers that cause various byte sizes for the tag information.
+    for field_number, tag_bytes in ((15, 1), (16, 2), (2047, 2), (2048, 3)):
+      expected_size = expected_value_size + tag_bytes
+      actual_size = byte_size_fn(field_number, value)
+      self.assertEqual(expected_size, actual_size,
+                       'byte_size_fn: %s, field_number: %d, value: %r\n'
+                       'Expected: %d, Actual: %d'% (
+          byte_size_fn, field_number, value, expected_size, actual_size))
+
+  def testByteSizeFunctions(self):
+    # Test all numeric *ByteSize() functions.
+    NUMERIC_ARGS = [
+        # Int32ByteSize().
+        [wire_format.Int32ByteSize, 0, 1],
+        [wire_format.Int32ByteSize, 127, 1],
+        [wire_format.Int32ByteSize, 128, 2],
+        [wire_format.Int32ByteSize, -1, 10],
+        # Int64ByteSize().
+        [wire_format.Int64ByteSize, 0, 1],
+        [wire_format.Int64ByteSize, 127, 1],
+        [wire_format.Int64ByteSize, 128, 2],
+        [wire_format.Int64ByteSize, -1, 10],
+        # UInt32ByteSize().
+        [wire_format.UInt32ByteSize, 0, 1],
+        [wire_format.UInt32ByteSize, 127, 1],
+        [wire_format.UInt32ByteSize, 128, 2],
+        [wire_format.UInt32ByteSize, wire_format.UINT32_MAX, 5],
+        # UInt64ByteSize().
+        [wire_format.UInt64ByteSize, 0, 1],
+        [wire_format.UInt64ByteSize, 127, 1],
+        [wire_format.UInt64ByteSize, 128, 2],
+        [wire_format.UInt64ByteSize, wire_format.UINT64_MAX, 10],
+        # SInt32ByteSize().
+        [wire_format.SInt32ByteSize, 0, 1],
+        [wire_format.SInt32ByteSize, -1, 1],
+        [wire_format.SInt32ByteSize, 1, 1],
+        [wire_format.SInt32ByteSize, -63, 1],
+        [wire_format.SInt32ByteSize, 63, 1],
+        [wire_format.SInt32ByteSize, -64, 1],
+        [wire_format.SInt32ByteSize, 64, 2],
+        # SInt64ByteSize().
+        [wire_format.SInt64ByteSize, 0, 1],
+        [wire_format.SInt64ByteSize, -1, 1],
+        [wire_format.SInt64ByteSize, 1, 1],
+        [wire_format.SInt64ByteSize, -63, 1],
+        [wire_format.SInt64ByteSize, 63, 1],
+        [wire_format.SInt64ByteSize, -64, 1],
+        [wire_format.SInt64ByteSize, 64, 2],
+        # Fixed32ByteSize().
+        [wire_format.Fixed32ByteSize, 0, 4],
+        [wire_format.Fixed32ByteSize, wire_format.UINT32_MAX, 4],
+        # Fixed64ByteSize().
+        [wire_format.Fixed64ByteSize, 0, 8],
+        [wire_format.Fixed64ByteSize, wire_format.UINT64_MAX, 8],
+        # SFixed32ByteSize().
+        [wire_format.SFixed32ByteSize, 0, 4],
+        [wire_format.SFixed32ByteSize, wire_format.INT32_MIN, 4],
+        [wire_format.SFixed32ByteSize, wire_format.INT32_MAX, 4],
+        # SFixed64ByteSize().
+        [wire_format.SFixed64ByteSize, 0, 8],
+        [wire_format.SFixed64ByteSize, wire_format.INT64_MIN, 8],
+        [wire_format.SFixed64ByteSize, wire_format.INT64_MAX, 8],
+        # FloatByteSize().
+        [wire_format.FloatByteSize, 0.0, 4],
+        [wire_format.FloatByteSize, 1000000000.0, 4],
+        [wire_format.FloatByteSize, -1000000000.0, 4],
+        # DoubleByteSize().
+        [wire_format.DoubleByteSize, 0.0, 8],
+        [wire_format.DoubleByteSize, 1000000000.0, 8],
+        [wire_format.DoubleByteSize, -1000000000.0, 8],
+        # BoolByteSize().
+        [wire_format.BoolByteSize, False, 1],
+        [wire_format.BoolByteSize, True, 1],
+        # EnumByteSize().
+        [wire_format.EnumByteSize, 0, 1],
+        [wire_format.EnumByteSize, 127, 1],
+        [wire_format.EnumByteSize, 128, 2],
+        [wire_format.EnumByteSize, wire_format.UINT32_MAX, 5],
+        ]
+    for args in NUMERIC_ARGS:
+      self.NumericByteSizeTestHelper(*args)
+
+    # Test strings and bytes.
+    for byte_size_fn in (wire_format.StringByteSize, wire_format.BytesByteSize):
+      # 1 byte for tag, 1 byte for length, 3 bytes for contents.
+      self.assertEqual(5, byte_size_fn(10, 'abc'))
+      # 2 bytes for tag, 1 byte for length, 3 bytes for contents.
+      self.assertEqual(6, byte_size_fn(16, 'abc'))
+      # 2 bytes for tag, 2 bytes for length, 128 bytes for contents.
+      self.assertEqual(132, byte_size_fn(16, 'a' * 128))
+
+    # Test UTF-8 string byte size calculation.
+    # 1 byte for tag, 1 byte for length, 8 bytes for content.
+    self.assertEqual(10, wire_format.StringByteSize(
+        5, unicode('\xd0\xa2\xd0\xb5\xd1\x81\xd1\x82', 'utf-8')))
+
+    class MockMessage(object):
+      def __init__(self, byte_size):
+        self.byte_size = byte_size
+      def ByteSize(self):
+        return self.byte_size
+
+    message_byte_size = 10
+    mock_message = MockMessage(byte_size=message_byte_size)
+    # Test groups.
+    # (2 * 1) bytes for begin and end tags, plus message_byte_size.
+    self.assertEqual(2 + message_byte_size,
+                     wire_format.GroupByteSize(1, mock_message))
+    # (2 * 2) bytes for begin and end tags, plus message_byte_size.
+    self.assertEqual(4 + message_byte_size,
+                     wire_format.GroupByteSize(16, mock_message))
+
+    # Test messages.
+    # 1 byte for tag, plus 1 byte for length, plus contents.
+    self.assertEqual(2 + mock_message.byte_size,
+                     wire_format.MessageByteSize(1, mock_message))
+    # 2 bytes for tag, plus 1 byte for length, plus contents.
+    self.assertEqual(3 + mock_message.byte_size,
+                     wire_format.MessageByteSize(16, mock_message))
+    # 2 bytes for tag, plus 2 bytes for length, plus contents.
+    mock_message.byte_size = 128
+    self.assertEqual(4 + mock_message.byte_size,
+                     wire_format.MessageByteSize(16, mock_message))
+
+
+    # Test message set item byte size.
+    # 4 bytes for tags, plus 1 byte for length, plus 1 byte for type_id,
+    # plus contents.
+    mock_message.byte_size = 10
+    self.assertEqual(mock_message.byte_size + 6,
+                     wire_format.MessageSetItemByteSize(1, mock_message))
+
+    # 4 bytes for tags, plus 2 bytes for length, plus 1 byte for type_id,
+    # plus contents.
+    mock_message.byte_size = 128
+    self.assertEqual(mock_message.byte_size + 7,
+                     wire_format.MessageSetItemByteSize(1, mock_message))
+
+    # 4 bytes for tags, plus 2 bytes for length, plus 2 byte for type_id,
+    # plus contents.
+    self.assertEqual(mock_message.byte_size + 8,
+                     wire_format.MessageSetItemByteSize(128, mock_message))
+
+    # Too-long varint.
+    self.assertRaises(message.EncodeError,
+                      wire_format.UInt64ByteSize, 1, 1 << 128)
+
+
+if __name__ == '__main__':
+  unittest.main()