change metaclass of Layer from pybind11_builtins.pybind11_type to type

[zhiqiu]

PR types

New features

PR changes

Others

Describe

change metaclass of Layer from pybind11_builtins.pybind11_type to type

By default, the mateclass of Layer is pybind11_builtins.pybind11_type, and the mateclass of pybind11_builtins.pybind11_type is type.
This PR changes mateclass of Layer to type directly to fix some problem.

For example,

import paddle
import cloudpickle
import os
import subprocess
import torch

class PaddleLayer(paddle.nn.Layer):
    def __init__(self):
        super(PaddleLayer, self).__init__()
        self.linear_1 = paddle.nn.Linear(784, 512)

class TorchModule(torch.nn.Module):
    def __init__(self):
        super(TorchModule, self).__init__()
        self.conv1 = torch.nn.Conv2d(1, 20, 5)

class OtherClass(object):
    def __init__(self):
        pass

def test_cloudpickle(cls):
    def to_str(byte):
        """ convert byte to string in pytohn2/3
        """
        return str(byte.decode())

    encoded = cloudpickle.dumps(cls)
    decoded_cls = cloudpickle.loads(encoded) # can be deserialized in the same process
    obj = decoded_cls()

    fname = "{}.cloudpickle".format(cls.__name__)
    with open(fname, 'wb') as f:
        f.write(encoded)

    command = """python -c '
import cloudpickle
with open("{}", "rb") as f:
    encoded = f.read()
    decoded_cls = cloudpickle.loads(encoded)
    obj = decoded_cls()
'
""".format(fname)
    # cannot be deserialized in another process
    try:
        subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT)
        print("load {} ok".format(fname))
    except subprocess.CalledProcessError as e:
        print("load {} failed".format(fname))
        print(str(e.output.decode()))

test_cloudpickle(OtherClass)
test_cloudpickle(TorchModule)
test_cloudpickle(PaddleLayer)

• before

load OtherClass.cloudpickle ok
load TorchModule.cloudpickle ok
load PaddleLayer.cloudpickle failed
Traceback (most recent call last):
  File "<string>", line 5, in <module>
  File "/usr/local/lib/python3.7/dist-packages/cloudpickle/cloudpickle.py", line 736, in _make_skeleton_class
    lambda ns: ns.update(type_kwargs)
  File "/usr/lib/python3.7/types.py", line 65, in new_class
    meta, ns, kwds = prepare_class(name, resolved_bases, kwds)
  File "/usr/lib/python3.7/types.py", line 118, in prepare_class
    meta = _calculate_meta(meta, bases)
  File "/usr/lib/python3.7/types.py", line 136, in _calculate_meta
    raise TypeError("metaclass conflict: "
TypeError: metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases

• after

load OtherClass.cloudpickle ok
load TorchModule.cloudpickle ok
load PaddleLayer.cloudpickle ok

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[chenwhql]

LGTM