[SPARK-29536][PYTHON] Upgrade cloudpickle to 1.1.1 to support Python 3.8

### What changes were proposed in this pull request? Inline cloudpickle in PySpark to cloudpickle 1.1.1. See https://github.com/cloudpipe/cloudpickle/blob/v1.1.1/cloudpickle/cloudpickle.py cloudpipe/cloudpickle#269 was added for Python 3.8 support (fixed from 1.1.0). Using 1.2.2 seems breaking PyPy 2 due to cloudpipe/cloudpickle#278 so this PR currently uses 1.1.1. Once we drop Python 2, we can switch to the highest version. ### Why are the changes needed? positional-only arguments was newly introduced from Python 3.8 (see https://docs.python.org/3/whatsnew/3.8.html#positional-only-parameters) Particularly the newly added argument to `types.CodeType` was the problem (https://docs.python.org/3/whatsnew/3.8.html#changes-in-the-python-api): > `types.CodeType` has a new parameter in the second position of the constructor (posonlyargcount) to support positional-only arguments defined in **PEP 570**. The first argument (argcount) now represents the total number of positional arguments (including positional-only arguments). The new `replace()` method of `types.CodeType` can be used to make the code future-proof. ### Does this PR introduce any user-facing change? No. ### How was this patch tested? Manually tested. Note that the optional dependency PyArrow looks not yet supporting Python 3.8; therefore, it was not tested. See "Details" below. <details> <p> ```bash cd python ./run-tests --python-executables=python3.8 ``` ``` Running PySpark tests. Output is in /Users/hyukjin.kwon/workspace/forked/spark/python/unit-tests.log Will test against the following Python executables: ['python3.8'] Will test the following Python modules: ['pyspark-core', 'pyspark-ml', 'pyspark-mllib', 'pyspark-sql', 'pyspark-streaming'] Starting test(python3.8): pyspark.ml.tests.test_algorithms Starting test(python3.8): pyspark.ml.tests.test_feature Starting test(python3.8): pyspark.ml.tests.test_base Starting test(python3.8): pyspark.ml.tests.test_evaluation Finished test(python3.8): pyspark.ml.tests.test_base (12s) Starting test(python3.8): pyspark.ml.tests.test_image Finished test(python3.8): pyspark.ml.tests.test_evaluation (14s) Starting test(python3.8): pyspark.ml.tests.test_linalg Finished test(python3.8): pyspark.ml.tests.test_feature (23s) Starting test(python3.8): pyspark.ml.tests.test_param Finished test(python3.8): pyspark.ml.tests.test_image (22s) Starting test(python3.8): pyspark.ml.tests.test_persistence Finished test(python3.8): pyspark.ml.tests.test_param (25s) Starting test(python3.8): pyspark.ml.tests.test_pipeline Finished test(python3.8): pyspark.ml.tests.test_linalg (37s) Starting test(python3.8): pyspark.ml.tests.test_stat Finished test(python3.8): pyspark.ml.tests.test_pipeline (7s) Starting test(python3.8): pyspark.ml.tests.test_training_summary Finished test(python3.8): pyspark.ml.tests.test_stat (21s) Starting test(python3.8): pyspark.ml.tests.test_tuning Finished test(python3.8): pyspark.ml.tests.test_persistence (45s) Starting test(python3.8): pyspark.ml.tests.test_wrapper Finished test(python3.8): pyspark.ml.tests.test_algorithms (83s) Starting test(python3.8): pyspark.mllib.tests.test_algorithms Finished test(python3.8): pyspark.ml.tests.test_training_summary (32s) Starting test(python3.8): pyspark.mllib.tests.test_feature Finished test(python3.8): pyspark.ml.tests.test_wrapper (20s) Starting test(python3.8): pyspark.mllib.tests.test_linalg Finished test(python3.8): pyspark.mllib.tests.test_feature (32s) Starting test(python3.8): pyspark.mllib.tests.test_stat Finished test(python3.8): pyspark.mllib.tests.test_algorithms (70s) Starting test(python3.8): pyspark.mllib.tests.test_streaming_algorithms Finished test(python3.8): pyspark.mllib.tests.test_stat (37s) Starting test(python3.8): pyspark.mllib.tests.test_util Finished test(python3.8): pyspark.mllib.tests.test_linalg (70s) Starting test(python3.8): pyspark.sql.tests.test_arrow Finished test(python3.8): pyspark.sql.tests.test_arrow (1s) ... 53 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_catalog Finished test(python3.8): pyspark.mllib.tests.test_util (15s) Starting test(python3.8): pyspark.sql.tests.test_column Finished test(python3.8): pyspark.sql.tests.test_catalog (24s) Starting test(python3.8): pyspark.sql.tests.test_conf Finished test(python3.8): pyspark.sql.tests.test_column (21s) Starting test(python3.8): pyspark.sql.tests.test_context Finished test(python3.8): pyspark.ml.tests.test_tuning (125s) Starting test(python3.8): pyspark.sql.tests.test_dataframe Finished test(python3.8): pyspark.sql.tests.test_conf (9s) Starting test(python3.8): pyspark.sql.tests.test_datasources Finished test(python3.8): pyspark.sql.tests.test_context (29s) Starting test(python3.8): pyspark.sql.tests.test_functions Finished test(python3.8): pyspark.sql.tests.test_datasources (32s) Starting test(python3.8): pyspark.sql.tests.test_group Finished test(python3.8): pyspark.sql.tests.test_dataframe (39s) ... 3 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf Finished test(python3.8): pyspark.sql.tests.test_pandas_udf (1s) ... 6 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf_cogrouped_map Finished test(python3.8): pyspark.sql.tests.test_pandas_udf_cogrouped_map (0s) ... 14 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf_grouped_agg Finished test(python3.8): pyspark.sql.tests.test_pandas_udf_grouped_agg (1s) ... 15 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf_grouped_map Finished test(python3.8): pyspark.sql.tests.test_pandas_udf_grouped_map (1s) ... 20 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf_scalar Finished test(python3.8): pyspark.sql.tests.test_pandas_udf_scalar (1s) ... 49 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_pandas_udf_window Finished test(python3.8): pyspark.sql.tests.test_pandas_udf_window (1s) ... 14 tests were skipped Starting test(python3.8): pyspark.sql.tests.test_readwriter Finished test(python3.8): pyspark.sql.tests.test_functions (29s) Starting test(python3.8): pyspark.sql.tests.test_serde Finished test(python3.8): pyspark.sql.tests.test_group (20s) Starting test(python3.8): pyspark.sql.tests.test_session Finished test(python3.8): pyspark.mllib.tests.test_streaming_algorithms (126s) Starting test(python3.8): pyspark.sql.tests.test_streaming Finished test(python3.8): pyspark.sql.tests.test_serde (25s) Starting test(python3.8): pyspark.sql.tests.test_types Finished test(python3.8): pyspark.sql.tests.test_readwriter (38s) Starting test(python3.8): pyspark.sql.tests.test_udf Finished test(python3.8): pyspark.sql.tests.test_session (32s) Starting test(python3.8): pyspark.sql.tests.test_utils Finished test(python3.8): pyspark.sql.tests.test_utils (17s) Starting test(python3.8): pyspark.streaming.tests.test_context Finished test(python3.8): pyspark.sql.tests.test_types (45s) Starting test(python3.8): pyspark.streaming.tests.test_dstream Finished test(python3.8): pyspark.sql.tests.test_udf (44s) Starting test(python3.8): pyspark.streaming.tests.test_kinesis Finished test(python3.8): pyspark.streaming.tests.test_kinesis (0s) ... 2 tests were skipped Starting test(python3.8): pyspark.streaming.tests.test_listener Finished test(python3.8): pyspark.streaming.tests.test_context (28s) Starting test(python3.8): pyspark.tests.test_appsubmit Finished test(python3.8): pyspark.sql.tests.test_streaming (60s) Starting test(python3.8): pyspark.tests.test_broadcast Finished test(python3.8): pyspark.streaming.tests.test_listener (11s) Starting test(python3.8): pyspark.tests.test_conf Finished test(python3.8): pyspark.tests.test_conf (17s) Starting test(python3.8): pyspark.tests.test_context Finished test(python3.8): pyspark.tests.test_broadcast (39s) Starting test(python3.8): pyspark.tests.test_daemon Finished test(python3.8): pyspark.tests.test_daemon (5s) Starting test(python3.8): pyspark.tests.test_join Finished test(python3.8): pyspark.tests.test_context (31s) Starting test(python3.8): pyspark.tests.test_profiler Finished test(python3.8): pyspark.tests.test_join (9s) Starting test(python3.8): pyspark.tests.test_rdd Finished test(python3.8): pyspark.tests.test_profiler (12s) Starting test(python3.8): pyspark.tests.test_readwrite Finished test(python3.8): pyspark.tests.test_readwrite (23s) ... 3 tests were skipped Starting test(python3.8): pyspark.tests.test_serializers Finished test(python3.8): pyspark.tests.test_appsubmit (94s) Starting test(python3.8): pyspark.tests.test_shuffle Finished test(python3.8): pyspark.streaming.tests.test_dstream (110s) Starting test(python3.8): pyspark.tests.test_taskcontext Finished test(python3.8): pyspark.tests.test_rdd (42s) Starting test(python3.8): pyspark.tests.test_util Finished test(python3.8): pyspark.tests.test_serializers (11s) Starting test(python3.8): pyspark.tests.test_worker Finished test(python3.8): pyspark.tests.test_shuffle (12s) Starting test(python3.8): pyspark.accumulators Finished test(python3.8): pyspark.tests.test_util (7s) Starting test(python3.8): pyspark.broadcast Finished test(python3.8): pyspark.accumulators (8s) Starting test(python3.8): pyspark.conf Finished test(python3.8): pyspark.broadcast (8s) Starting test(python3.8): pyspark.context Finished test(python3.8): pyspark.tests.test_worker (19s) Starting test(python3.8): pyspark.ml.classification Finished test(python3.8): pyspark.conf (4s) Starting test(python3.8): pyspark.ml.clustering Finished test(python3.8): pyspark.context (22s) Starting test(python3.8): pyspark.ml.evaluation Finished test(python3.8): pyspark.tests.test_taskcontext (49s) Starting test(python3.8): pyspark.ml.feature Finished test(python3.8): pyspark.ml.clustering (43s) Starting test(python3.8): pyspark.ml.fpm Finished test(python3.8): pyspark.ml.evaluation (27s) Starting test(python3.8): pyspark.ml.image Finished test(python3.8): pyspark.ml.image (8s) Starting test(python3.8): pyspark.ml.linalg.__init__ Finished test(python3.8): pyspark.ml.linalg.__init__ (0s) Starting test(python3.8): pyspark.ml.recommendation Finished test(python3.8): pyspark.ml.classification (63s) Starting test(python3.8): pyspark.ml.regression Finished test(python3.8): pyspark.ml.fpm (23s) Starting test(python3.8): pyspark.ml.stat Finished test(python3.8): pyspark.ml.stat (30s) Starting test(python3.8): pyspark.ml.tuning Finished test(python3.8): pyspark.ml.regression (51s) Starting test(python3.8): pyspark.mllib.classification Finished test(python3.8): pyspark.ml.feature (93s) Starting test(python3.8): pyspark.mllib.clustering Finished test(python3.8): pyspark.ml.tuning (39s) Starting test(python3.8): pyspark.mllib.evaluation Finished test(python3.8): pyspark.mllib.classification (38s) Starting test(python3.8): pyspark.mllib.feature Finished test(python3.8): pyspark.mllib.evaluation (25s) Starting test(python3.8): pyspark.mllib.fpm Finished test(python3.8): pyspark.mllib.clustering (64s) Starting test(python3.8): pyspark.mllib.linalg.__init__ Finished test(python3.8): pyspark.ml.recommendation (131s) Starting test(python3.8): pyspark.mllib.linalg.distributed Finished test(python3.8): pyspark.mllib.linalg.__init__ (0s) Starting test(python3.8): pyspark.mllib.random Finished test(python3.8): pyspark.mllib.feature (36s) Starting test(python3.8): pyspark.mllib.recommendation Finished test(python3.8): pyspark.mllib.fpm (31s) Starting test(python3.8): pyspark.mllib.regression Finished test(python3.8): pyspark.mllib.random (16s) Starting test(python3.8): pyspark.mllib.stat.KernelDensity Finished test(python3.8): pyspark.mllib.stat.KernelDensity (1s) Starting test(python3.8): pyspark.mllib.stat._statistics Finished test(python3.8): pyspark.mllib.stat._statistics (25s) Starting test(python3.8): pyspark.mllib.tree Finished test(python3.8): pyspark.mllib.regression (44s) Starting test(python3.8): pyspark.mllib.util Finished test(python3.8): pyspark.mllib.recommendation (49s) Starting test(python3.8): pyspark.profiler Finished test(python3.8): pyspark.mllib.linalg.distributed (53s) Starting test(python3.8): pyspark.rdd Finished test(python3.8): pyspark.profiler (14s) Starting test(python3.8): pyspark.serializers Finished test(python3.8): pyspark.mllib.tree (30s) Starting test(python3.8): pyspark.shuffle Finished test(python3.8): pyspark.shuffle (2s) Starting test(python3.8): pyspark.sql.avro.functions Finished test(python3.8): pyspark.mllib.util (30s) Starting test(python3.8): pyspark.sql.catalog Finished test(python3.8): pyspark.serializers (17s) Starting test(python3.8): pyspark.sql.column Finished test(python3.8): pyspark.rdd (31s) Starting test(python3.8): pyspark.sql.conf Finished test(python3.8): pyspark.sql.conf (7s) Starting test(python3.8): pyspark.sql.context Finished test(python3.8): pyspark.sql.avro.functions (19s) Starting test(python3.8): pyspark.sql.dataframe Finished test(python3.8): pyspark.sql.catalog (16s) Starting test(python3.8): pyspark.sql.functions Finished test(python3.8): pyspark.sql.column (27s) Starting test(python3.8): pyspark.sql.group Finished test(python3.8): pyspark.sql.context (26s) Starting test(python3.8): pyspark.sql.readwriter Finished test(python3.8): pyspark.sql.group (52s) Starting test(python3.8): pyspark.sql.session Finished test(python3.8): pyspark.sql.dataframe (73s) Starting test(python3.8): pyspark.sql.streaming Finished test(python3.8): pyspark.sql.functions (75s) Starting test(python3.8): pyspark.sql.types Finished test(python3.8): pyspark.sql.readwriter (57s) Starting test(python3.8): pyspark.sql.udf Finished test(python3.8): pyspark.sql.types (13s) Starting test(python3.8): pyspark.sql.window Finished test(python3.8): pyspark.sql.session (32s) Starting test(python3.8): pyspark.streaming.util Finished test(python3.8): pyspark.streaming.util (1s) Starting test(python3.8): pyspark.util Finished test(python3.8): pyspark.util (0s) Finished test(python3.8): pyspark.sql.streaming (30s) Finished test(python3.8): pyspark.sql.udf (27s) Finished test(python3.8): pyspark.sql.window (22s) Tests passed in 855 seconds ``` </p> </details> Closes apache#26194 from HyukjinKwon/SPARK-29536. Authored-by: HyukjinKwon <[email protected]> Signed-off-by: HyukjinKwon <[email protected]>
Deegue · Deegue · Oct 28, 2019 · Oct 22, 2019 · Oct 22, 2019 · Oct 22, 2019
commit 811d563fbf60203377e8462e4fad271c1140b4fa
diff --git a/python/pyspark/cloudpickle.py b/python/pyspark/cloudpickle.py
@@ -44,7 +44,6 @@
 
 import dis
 from functools import partial
-import importlib
 import io
 import itertools
 import logging
@@ -56,12 +55,26 @@
 import traceback
 import types
 import weakref
+import uuid
+import threading
+
+
+try:
+    from enum import Enum
+except ImportError:
+    Enum = None
 
 # cloudpickle is meant for inter process communication: we expect all
 # communicating processes to run the same Python version hence we favor
 # communication speed over compatibility:
 DEFAULT_PROTOCOL = pickle.HIGHEST_PROTOCOL
 
+# Track the provenance of reconstructed dynamic classes to make it possible to
+# recontruct instances from the matching singleton class definition when
+# appropriate and preserve the usual "isinstance" semantics of Python objects.
+_DYNAMIC_CLASS_TRACKER_BY_CLASS = weakref.WeakKeyDictionary()
+_DYNAMIC_CLASS_TRACKER_BY_ID = weakref.WeakValueDictionary()
+_DYNAMIC_CLASS_TRACKER_LOCK = threading.Lock()
 
 if sys.version_info[0] < 3:  # pragma: no branch
     from pickle import Pickler
@@ -71,12 +84,37 @@
         from StringIO import StringIO
     string_types = (basestring,)  # noqa
     PY3 = False
+    PY2 = True
+    PY2_WRAPPER_DESCRIPTOR_TYPE = type(object.__init__)
+    PY2_METHOD_WRAPPER_TYPE = type(object.__eq__)
+    PY2_CLASS_DICT_BLACKLIST = (PY2_METHOD_WRAPPER_TYPE,
+                                PY2_WRAPPER_DESCRIPTOR_TYPE)
 else:
     types.ClassType = type
     from pickle import _Pickler as Pickler
     from io import BytesIO as StringIO
     string_types = (str,)
     PY3 = True
+    PY2 = False
+
+
+def _ensure_tracking(class_def):
+    with _DYNAMIC_CLASS_TRACKER_LOCK:
+        class_tracker_id = _DYNAMIC_CLASS_TRACKER_BY_CLASS.get(class_def)
+        if class_tracker_id is None:
+            class_tracker_id = uuid.uuid4().hex
+            _DYNAMIC_CLASS_TRACKER_BY_CLASS[class_def] = class_tracker_id
+            _DYNAMIC_CLASS_TRACKER_BY_ID[class_tracker_id] = class_def
+    return class_tracker_id
+
+
+def _lookup_class_or_track(class_tracker_id, class_def):
+    if class_tracker_id is not None:
+        with _DYNAMIC_CLASS_TRACKER_LOCK:
+            class_def = _DYNAMIC_CLASS_TRACKER_BY_ID.setdefault(
+                class_tracker_id, class_def)
+            _DYNAMIC_CLASS_TRACKER_BY_CLASS[class_def] = class_tracker_id
+    return class_def
 
 
 def _make_cell_set_template_code():
@@ -112,7 +150,7 @@ def inner(value):
     # NOTE: we are marking the cell variable as a free variable intentionally
     # so that we simulate an inner function instead of the outer function. This
     # is what gives us the ``nonlocal`` behavior in a Python 2 compatible way.
-    if not PY3:  # pragma: no branch
+    if PY2:  # pragma: no branch
         return types.CodeType(
             co.co_argcount,
             co.co_nlocals,
@@ -130,24 +168,43 @@ def inner(value):
             (),
         )
     else:
-        return types.CodeType(
-            co.co_argcount,
-            co.co_kwonlyargcount,
-            co.co_nlocals,
-            co.co_stacksize,
-            co.co_flags,
-            co.co_code,
-            co.co_consts,
-            co.co_names,
-            co.co_varnames,
-            co.co_filename,
-            co.co_name,
-            co.co_firstlineno,
-            co.co_lnotab,
-            co.co_cellvars,  # this is the trickery
-            (),
-        )
-
+        if hasattr(types.CodeType, "co_posonlyargcount"):  # pragma: no branch
+            return types.CodeType(
+                co.co_argcount,
+                co.co_posonlyargcount,  # Python3.8 with PEP570
+                co.co_kwonlyargcount,
+                co.co_nlocals,
+                co.co_stacksize,
+                co.co_flags,
+                co.co_code,
+                co.co_consts,
+                co.co_names,
+                co.co_varnames,
+                co.co_filename,
+                co.co_name,
+                co.co_firstlineno,
+                co.co_lnotab,
+                co.co_cellvars,  # this is the trickery
+                (),
+            )
+        else:
+            return types.CodeType(
+                co.co_argcount,
+                co.co_kwonlyargcount,
+                co.co_nlocals,
+                co.co_stacksize,
+                co.co_flags,
+                co.co_code,
+                co.co_consts,
+                co.co_names,
+                co.co_varnames,
+                co.co_filename,
+                co.co_name,
+                co.co_firstlineno,
+                co.co_lnotab,
+                co.co_cellvars,  # this is the trickery
+                (),
+            )
 
 _cell_set_template_code = _make_cell_set_template_code()
 
@@ -220,7 +277,7 @@ def _walk_global_ops(code):
         global-referencing instructions in *code*.
         """
         code = getattr(code, 'co_code', b'')
-        if not PY3:  # pragma: no branch
+        if PY2:  # pragma: no branch
             code = map(ord, code)
 
         n = len(code)
@@ -250,6 +307,39 @@ def _walk_global_ops(code):
                 yield op, instr.arg
 
 
+def _extract_class_dict(cls):
+    """Retrieve a copy of the dict of a class without the inherited methods"""
+    clsdict = dict(cls.__dict__)  # copy dict proxy to a dict
+    if len(cls.__bases__) == 1:
+        inherited_dict = cls.__bases__[0].__dict__
+    else:
+        inherited_dict = {}
+        for base in reversed(cls.__bases__):
+            inherited_dict.update(base.__dict__)
+    to_remove = []
+    for name, value in clsdict.items():
+        try:
+            base_value = inherited_dict[name]
+            if value is base_value:
+                to_remove.append(name)
+            elif PY2:
+                # backward compat for Python 2
+                if hasattr(value, "im_func"):
+                    if value.im_func is getattr(base_value, "im_func", None):
+                        to_remove.append(name)
+                elif isinstance(value, PY2_CLASS_DICT_BLACKLIST):
+                    # On Python 2 we have no way to pickle those specific
+                    # methods types nor to check that they are actually
+                    # inherited. So we assume that they are always inherited
+                    # from builtin types.
+                    to_remove.append(name)
+        except KeyError:
+            pass
+    for name in to_remove:
+        clsdict.pop(name)
+    return clsdict
+
+
 class CloudPickler(Pickler):
 
     dispatch = Pickler.dispatch.copy()
@@ -277,7 +367,7 @@ def save_memoryview(self, obj):
 
     dispatch[memoryview] = save_memoryview
 
-    if not PY3:  # pragma: no branch
+    if PY2:  # pragma: no branch
         def save_buffer(self, obj):
             self.save(str(obj))
 
@@ -300,12 +390,23 @@ def save_codeobject(self, obj):
         Save a code object
         """
         if PY3:  # pragma: no branch
-            args = (
-                obj.co_argcount, obj.co_kwonlyargcount, obj.co_nlocals, obj.co_stacksize,
-                obj.co_flags, obj.co_code, obj.co_consts, obj.co_names, obj.co_varnames,
-                obj.co_filename, obj.co_name, obj.co_firstlineno, obj.co_lnotab, obj.co_freevars,
-                obj.co_cellvars
-            )
+            if hasattr(obj, "co_posonlyargcount"):  # pragma: no branch
+                args = (
+                    obj.co_argcount, obj.co_posonlyargcount,
+                    obj.co_kwonlyargcount, obj.co_nlocals, obj.co_stacksize,
+                    obj.co_flags, obj.co_code, obj.co_consts, obj.co_names,
+                    obj.co_varnames, obj.co_filename, obj.co_name,
+                    obj.co_firstlineno, obj.co_lnotab, obj.co_freevars,
+                    obj.co_cellvars
+                )
+            else:
+                args = (
+                    obj.co_argcount, obj.co_kwonlyargcount, obj.co_nlocals,
+                    obj.co_stacksize, obj.co_flags, obj.co_code, obj.co_consts,
+                    obj.co_names, obj.co_varnames, obj.co_filename,
+                    obj.co_name, obj.co_firstlineno, obj.co_lnotab,
+                    obj.co_freevars, obj.co_cellvars
+                )
         else:
             args = (
                 obj.co_argcount, obj.co_nlocals, obj.co_stacksize, obj.co_flags, obj.co_code,
@@ -460,15 +561,40 @@ def func():
                             # then discards the reference to it
                             self.write(pickle.POP)
 
-    def save_dynamic_class(self, obj):
+    def _save_dynamic_enum(self, obj, clsdict):
+        """Special handling for dynamic Enum subclasses
+
+        Use a dedicated Enum constructor (inspired by EnumMeta.__call__) as the
+        EnumMeta metaclass has complex initialization that makes the Enum
+        subclasses hold references to their own instances.
         """
-        Save a class that can't be stored as module global.
+        members = dict((e.name, e.value) for e in obj)
+
+        # Python 2.7 with enum34 can have no qualname:
+        qualname = getattr(obj, "__qualname__", None)
+
+        self.save_reduce(_make_skeleton_enum,
+                         (obj.__bases__, obj.__name__, qualname, members,
+                          obj.__module__, _ensure_tracking(obj), None),
+                         obj=obj)
+
+        # Cleanup the clsdict that will be passed to _rehydrate_skeleton_class:
+        # Those attributes are already handled by the metaclass.
+        for attrname in ["_generate_next_value_", "_member_names_",
+                         "_member_map_", "_member_type_",
+                         "_value2member_map_"]:
+            clsdict.pop(attrname, None)
+        for member in members:
+            clsdict.pop(member)
+
+    def save_dynamic_class(self, obj):
+        """Save a class that can't be stored as module global.
 
         This method is used to serialize classes that are defined inside
         functions, or that otherwise can't be serialized as attribute lookups
         from global modules.
         """
-        clsdict = dict(obj.__dict__)  # copy dict proxy to a dict
+        clsdict = _extract_class_dict(obj)
         clsdict.pop('__weakref__', None)
 
         # For ABCMeta in python3.7+, remove _abc_impl as it is not picklable.
@@ -496,8 +622,8 @@ def save_dynamic_class(self, obj):
                 for k in obj.__slots__:
                     clsdict.pop(k, None)
 
-        # If type overrides __dict__ as a property, include it in the type kwargs.
-        # In Python 2, we can't set this attribute after construction.
+        # If type overrides __dict__ as a property, include it in the type
+        # kwargs. In Python 2, we can't set this attribute after construction.
         __dict__ = clsdict.pop('__dict__', None)
         if isinstance(__dict__, property):
             type_kwargs['__dict__'] = __dict__
@@ -524,8 +650,16 @@ def save_dynamic_class(self, obj):
         write(pickle.MARK)
 
         # Create and memoize an skeleton class with obj's name and bases.
-        tp = type(obj)
-        self.save_reduce(tp, (obj.__name__, obj.__bases__, type_kwargs), obj=obj)
+        if Enum is not None and issubclass(obj, Enum):
+            # Special handling of Enum subclasses
+            self._save_dynamic_enum(obj, clsdict)
+        else:
+            # "Regular" class definition:
+            tp = type(obj)
+            self.save_reduce(_make_skeleton_class,
+                             (tp, obj.__name__, obj.__bases__, type_kwargs,
+                              _ensure_tracking(obj), None),
+                             obj=obj)
 
         # Now save the rest of obj's __dict__. Any references to obj
         # encountered while saving will point to the skeleton class.
@@ -778,7 +912,7 @@ def save_inst(self, obj):
         save(stuff)
         write(pickle.BUILD)
 
-    if not PY3:  # pragma: no branch
+    if PY2:  # pragma: no branch
         dispatch[types.InstanceType] = save_inst
 
     def save_property(self, obj):
@@ -1119,6 +1253,22 @@ def _make_skel_func(code, cell_count, base_globals=None):
     return types.FunctionType(code, base_globals, None, None, closure)
 
 
+def _make_skeleton_class(type_constructor, name, bases, type_kwargs,
+                         class_tracker_id, extra):
+    """Build dynamic class with an empty __dict__ to be filled once memoized
+
+    If class_tracker_id is not None, try to lookup an existing class definition
+    matching that id. If none is found, track a newly reconstructed class
+    definition under that id so that other instances stemming from the same
+    class id will also reuse this class definition.
+
+    The "extra" variable is meant to be a dict (or None) that can be used for
+    forward compatibility shall the need arise.
+    """
+    skeleton_class = type_constructor(name, bases, type_kwargs)
+    return _lookup_class_or_track(class_tracker_id, skeleton_class)
+
+
 def _rehydrate_skeleton_class(skeleton_class, class_dict):
     """Put attributes from `class_dict` back on `skeleton_class`.
 
@@ -1137,6 +1287,39 @@ def _rehydrate_skeleton_class(skeleton_class, class_dict):
     return skeleton_class
 
 
+def _make_skeleton_enum(bases, name, qualname, members, module,
+                        class_tracker_id, extra):
+    """Build dynamic enum with an empty __dict__ to be filled once memoized
+
+    The creation of the enum class is inspired by the code of
+    EnumMeta._create_.
+
+    If class_tracker_id is not None, try to lookup an existing enum definition
+    matching that id. If none is found, track a newly reconstructed enum
+    definition under that id so that other instances stemming from the same
+    class id will also reuse this enum definition.
+
+    The "extra" variable is meant to be a dict (or None) that can be used for
+    forward compatibility shall the need arise.
+    """
+    # enums always inherit from their base Enum class at the last position in
+    # the list of base classes:
+    enum_base = bases[-1]
+    metacls = enum_base.__class__
+    classdict = metacls.__prepare__(name, bases)
+
+    for member_name, member_value in members.items():
+        classdict[member_name] = member_value
+    enum_class = metacls.__new__(metacls, name, bases, classdict)
+    enum_class.__module__ = module
+
+    # Python 2.7 compat
+    if qualname is not None:
+        enum_class.__qualname__ = qualname
+
+    return _lookup_class_or_track(class_tracker_id, enum_class)
+
+
 def _is_dynamic(module):
     """
     Return True if the module is special module that cannot be imported by its
@@ -1176,4 +1359,4 @@ def _reduce_method_descriptor(obj):
         import copy_reg as copyreg
     except ImportError:
         import copyreg
-    copyreg.pickle(method_descriptor, _reduce_method_descriptor)
+    copyreg.pickle(method_descriptor, _reduce_method_descriptor)
diff --git a/python/setup.py b/python/setup.py
@@ -230,6 +230,7 @@ def _supports_symlinks():
             'Programming Language :: Python :: 3.5',
             'Programming Language :: Python :: 3.6',
             'Programming Language :: Python :: 3.7',
+            'Programming Language :: Python :: 3.8',
             'Programming Language :: Python :: Implementation :: CPython',
             'Programming Language :: Python :: Implementation :: PyPy']
     )